YOLOv5 v6.0 compatibility update (#1857)

* Initial commit

* Initial commit

* Cleanup

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Fix precommit errors

* Remove TF builds from CI

* export last.pt

* Created using Colaboratory

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit is contained in:
Glenn Jocher
2021-11-14 22:22:59 +01:00
committed by GitHub
parent 1be31704c9
commit 7eb23e3c1d
90 changed files with 6642 additions and 4145 deletions
+18
View File
@@ -0,0 +1,18 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
utils/initialization
"""
def notebook_init():
# For notebooks
print('Checking setup...')
from IPython import display # to display images and clear console output
from utils.general import emojis
from utils.torch_utils import select_device # imports
display.clear_output()
select_device(newline=False)
print(emojis('Setup complete ✅'))
return display
+5 -2
View File
@@ -1,4 +1,7 @@
# Activation functions
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Activation functions
"""
import torch
import torch.nn as nn
@@ -16,7 +19,7 @@ class Hardswish(nn.Module): # export-friendly version of nn.Hardswish()
@staticmethod
def forward(x):
# return x * F.hardsigmoid(x) # for torchscript and CoreML
return x * F.hardtanh(x + 3, 0., 6.) / 6. # for torchscript, CoreML and ONNX
return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0 # for torchscript, CoreML and ONNX
# Mish https://github.com/digantamisra98/Mish --------------------------------------------------------------------------
+277
View File
@@ -0,0 +1,277 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Image augmentation functions
"""
import math
import random
import cv2
import numpy as np
from utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box
from utils.metrics import bbox_ioa
class Albumentations:
# Albumentations class (optional, only used if package is installed)
def __init__(self):
self.transform = None
try:
import albumentations as A
check_version(A.__version__, '1.0.3', hard=True) # version requirement
self.transform = A.Compose([
A.Blur(p=0.01),
A.MedianBlur(p=0.01),
A.ToGray(p=0.01),
A.CLAHE(p=0.01),
A.RandomBrightnessContrast(p=0.0),
A.RandomGamma(p=0.0),
A.ImageCompression(quality_lower=75, p=0.0)],
bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
LOGGER.info(colorstr('albumentations: ') + ', '.join(f'{x}' for x in self.transform.transforms if x.p))
except ImportError: # package not installed, skip
pass
except Exception as e:
LOGGER.info(colorstr('albumentations: ') + f'{e}')
def __call__(self, im, labels, p=1.0):
if self.transform and random.random() < p:
new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0]) # transformed
im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])])
return im, labels
def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
# HSV color-space augmentation
if hgain or sgain or vgain:
r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1 # random gains
hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV))
dtype = im.dtype # uint8
x = np.arange(0, 256, dtype=r.dtype)
lut_hue = ((x * r[0]) % 180).astype(dtype)
lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im) # no return needed
def hist_equalize(im, clahe=True, bgr=False):
# Equalize histogram on BGR image 'im' with im.shape(n,m,3) and range 0-255
yuv = cv2.cvtColor(im, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV)
if clahe:
c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
yuv[:, :, 0] = c.apply(yuv[:, :, 0])
else:
yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0]) # equalize Y channel histogram
return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB) # convert YUV image to RGB
def replicate(im, labels):
# Replicate labels
h, w = im.shape[:2]
boxes = labels[:, 1:].astype(int)
x1, y1, x2, y2 = boxes.T
s = ((x2 - x1) + (y2 - y1)) / 2 # side length (pixels)
for i in s.argsort()[:round(s.size * 0.5)]: # smallest indices
x1b, y1b, x2b, y2b = boxes[i]
bh, bw = y2b - y1b, x2b - x1b
yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw)) # offset x, y
x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
im[y1a:y2a, x1a:x2a] = im[y1b:y2b, x1b:x2b] # im4[ymin:ymax, xmin:xmax]
labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
return im, labels
def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
# Resize and pad image while meeting stride-multiple constraints
shape = im.shape[:2] # current shape [height, width]
if isinstance(new_shape, int):
new_shape = (new_shape, new_shape)
# Scale ratio (new / old)
r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
if not scaleup: # only scale down, do not scale up (for better val mAP)
r = min(r, 1.0)
# Compute padding
ratio = r, r # width, height ratios
new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1] # wh padding
if auto: # minimum rectangle
dw, dh = np.mod(dw, stride), np.mod(dh, stride) # wh padding
elif scaleFill: # stretch
dw, dh = 0.0, 0.0
new_unpad = (new_shape[1], new_shape[0])
ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # width, height ratios
dw /= 2 # divide padding into 2 sides
dh /= 2
if shape[::-1] != new_unpad: # resize
im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color) # add border
return im, ratio, (dw, dh)
def random_perspective(im, targets=(), segments=(), degrees=10, translate=.1, scale=.1, shear=10, perspective=0.0,
border=(0, 0)):
# torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(0.1, 0.1), scale=(0.9, 1.1), shear=(-10, 10))
# targets = [cls, xyxy]
height = im.shape[0] + border[0] * 2 # shape(h,w,c)
width = im.shape[1] + border[1] * 2
# Center
C = np.eye(3)
C[0, 2] = -im.shape[1] / 2 # x translation (pixels)
C[1, 2] = -im.shape[0] / 2 # y translation (pixels)
# Perspective
P = np.eye(3)
P[2, 0] = random.uniform(-perspective, perspective) # x perspective (about y)
P[2, 1] = random.uniform(-perspective, perspective) # y perspective (about x)
# Rotation and Scale
R = np.eye(3)
a = random.uniform(-degrees, degrees)
# a += random.choice([-180, -90, 0, 90]) # add 90deg rotations to small rotations
s = random.uniform(1 - scale, 1 + scale)
# s = 2 ** random.uniform(-scale, scale)
R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
# Shear
S = np.eye(3)
S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # x shear (deg)
S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # y shear (deg)
# Translation
T = np.eye(3)
T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width # x translation (pixels)
T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height # y translation (pixels)
# Combined rotation matrix
M = T @ S @ R @ P @ C # order of operations (right to left) is IMPORTANT
if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any(): # image changed
if perspective:
im = cv2.warpPerspective(im, M, dsize=(width, height), borderValue=(114, 114, 114))
else: # affine
im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
# Visualize
# import matplotlib.pyplot as plt
# ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
# ax[0].imshow(im[:, :, ::-1]) # base
# ax[1].imshow(im2[:, :, ::-1]) # warped
# Transform label coordinates
n = len(targets)
if n:
use_segments = any(x.any() for x in segments)
new = np.zeros((n, 4))
if use_segments: # warp segments
segments = resample_segments(segments) # upsample
for i, segment in enumerate(segments):
xy = np.ones((len(segment), 3))
xy[:, :2] = segment
xy = xy @ M.T # transform
xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2] # perspective rescale or affine
# clip
new[i] = segment2box(xy, width, height)
else: # warp boxes
xy = np.ones((n * 4, 3))
xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1
xy = xy @ M.T # transform
xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8) # perspective rescale or affine
# create new boxes
x = xy[:, [0, 2, 4, 6]]
y = xy[:, [1, 3, 5, 7]]
new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
# clip
new[:, [0, 2]] = new[:, [0, 2]].clip(0, width)
new[:, [1, 3]] = new[:, [1, 3]].clip(0, height)
# filter candidates
i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10)
targets = targets[i]
targets[:, 1:5] = new[i]
return im, targets
def copy_paste(im, labels, segments, p=0.5):
# Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy)
n = len(segments)
if p and n:
h, w, c = im.shape # height, width, channels
im_new = np.zeros(im.shape, np.uint8)
for j in random.sample(range(n), k=round(p * n)):
l, s = labels[j], segments[j]
box = w - l[3], l[2], w - l[1], l[4]
ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
if (ioa < 0.30).all(): # allow 30% obscuration of existing labels
labels = np.concatenate((labels, [[l[0], *box]]), 0)
segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1))
cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED)
result = cv2.bitwise_and(src1=im, src2=im_new)
result = cv2.flip(result, 1) # augment segments (flip left-right)
i = result > 0 # pixels to replace
# i[:, :] = result.max(2).reshape(h, w, 1) # act over ch
im[i] = result[i] # cv2.imwrite('debug.jpg', im) # debug
return im, labels, segments
def cutout(im, labels, p=0.5):
# Applies image cutout augmentation https://arxiv.org/abs/1708.04552
if random.random() < p:
h, w = im.shape[:2]
scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16 # image size fraction
for s in scales:
mask_h = random.randint(1, int(h * s)) # create random masks
mask_w = random.randint(1, int(w * s))
# box
xmin = max(0, random.randint(0, w) - mask_w // 2)
ymin = max(0, random.randint(0, h) - mask_h // 2)
xmax = min(w, xmin + mask_w)
ymax = min(h, ymin + mask_h)
# apply random color mask
im[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)]
# return unobscured labels
if len(labels) and s > 0.03:
box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
labels = labels[ioa < 0.60] # remove >60% obscured labels
return labels
def mixup(im, labels, im2, labels2):
# Applies MixUp augmentation https://arxiv.org/pdf/1710.09412.pdf
r = np.random.beta(32.0, 32.0) # mixup ratio, alpha=beta=32.0
im = (im * r + im2 * (1 - r)).astype(np.uint8)
labels = np.concatenate((labels, labels2), 0)
return im, labels
def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n)
# Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps)) # aspect ratio
return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr) # candidates
+47 -44
View File
@@ -1,28 +1,32 @@
# Auto-anchor utils
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Auto-anchor utils
"""
import random
import numpy as np
import torch
import yaml
from tqdm import tqdm
from utils.general import colorstr
from utils.general import LOGGER, colorstr, emojis
PREFIX = colorstr('AutoAnchor: ')
def check_anchor_order(m):
# Check anchor order against stride order for YOLOv3 Detect() module m, and correct if necessary
a = m.anchor_grid.prod(-1).view(-1) # anchor area
# Check anchor order against stride order for Detect() module m, and correct if necessary
a = m.anchors.prod(-1).view(-1) # anchor area
da = a[-1] - a[0] # delta a
ds = m.stride[-1] - m.stride[0] # delta s
if da.sign() != ds.sign(): # same order
print('Reversing anchor order')
LOGGER.info(f'{PREFIX}Reversing anchor order')
m.anchors[:] = m.anchors.flip(0)
m.anchor_grid[:] = m.anchor_grid.flip(0)
def check_anchors(dataset, model, thr=4.0, imgsz=640):
# Check anchor fit to data, recompute if necessary
prefix = colorstr('autoanchor: ')
print(f'\n{prefix}Analyzing anchors... ', end='')
m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect()
shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True)
scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1)) # augment scale
@@ -30,39 +34,39 @@ def check_anchors(dataset, model, thr=4.0, imgsz=640):
def metric(k): # compute metric
r = wh[:, None] / k[None]
x = torch.min(r, 1. / r).min(2)[0] # ratio metric
x = torch.min(r, 1 / r).min(2)[0] # ratio metric
best = x.max(1)[0] # best_x
aat = (x > 1. / thr).float().sum(1).mean() # anchors above threshold
bpr = (best > 1. / thr).float().mean() # best possible recall
aat = (x > 1 / thr).float().sum(1).mean() # anchors above threshold
bpr = (best > 1 / thr).float().mean() # best possible recall
return bpr, aat
anchors = m.anchor_grid.clone().cpu().view(-1, 2) # current anchors
bpr, aat = metric(anchors)
print(f'anchors/target = {aat:.2f}, Best Possible Recall (BPR) = {bpr:.4f}', end='')
if bpr < 0.98: # threshold to recompute
print('. Attempting to improve anchors, please wait...')
na = m.anchor_grid.numel() // 2 # number of anchors
anchors = m.anchors.clone() * m.stride.to(m.anchors.device).view(-1, 1, 1) # current anchors
bpr, aat = metric(anchors.cpu().view(-1, 2))
s = f'\n{PREFIX}{aat:.2f} anchors/target, {bpr:.3f} Best Possible Recall (BPR). '
if bpr > 0.98: # threshold to recompute
LOGGER.info(emojis(f'{s}Current anchors are a good fit to dataset ✅'))
else:
LOGGER.info(emojis(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...'))
na = m.anchors.numel() // 2 # number of anchors
try:
anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
except Exception as e:
print(f'{prefix}ERROR: {e}')
LOGGER.info(f'{PREFIX}ERROR: {e}')
new_bpr = metric(anchors)[0]
if new_bpr > bpr: # replace anchors
anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors)
m.anchor_grid[:] = anchors.clone().view_as(m.anchor_grid) # for inference
m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss
check_anchor_order(m)
print(f'{prefix}New anchors saved to model. Update model *.yaml to use these anchors in the future.')
LOGGER.info(f'{PREFIX}New anchors saved to model. Update model *.yaml to use these anchors in the future.')
else:
print(f'{prefix}Original anchors better than new anchors. Proceeding with original anchors.')
print('') # newline
LOGGER.info(f'{PREFIX}Original anchors better than new anchors. Proceeding with original anchors.')
def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
""" Creates kmeans-evolved anchors from training dataset
Arguments:
path: path to dataset *.yaml, or a loaded dataset
dataset: path to data.yaml, or a loaded dataset
n: number of anchors
img_size: image size used for training
thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0
@@ -77,12 +81,11 @@ def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=10
"""
from scipy.cluster.vq import kmeans
thr = 1. / thr
prefix = colorstr('autoanchor: ')
thr = 1 / thr
def metric(k, wh): # compute metrics
r = wh[:, None] / k[None]
x = torch.min(r, 1. / r).min(2)[0] # ratio metric
x = torch.min(r, 1 / r).min(2)[0] # ratio metric
# x = wh_iou(wh, torch.tensor(k)) # iou metric
return x, x.max(1)[0] # x, best_x
@@ -90,24 +93,24 @@ def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=10
_, best = metric(torch.tensor(k, dtype=torch.float32), wh)
return (best * (best > thr).float()).mean() # fitness
def print_results(k):
def print_results(k, verbose=True):
k = k[np.argsort(k.prod(1))] # sort small to large
x, best = metric(k, wh0)
bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n # best possible recall, anch > thr
print(f'{prefix}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr')
print(f'{prefix}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, '
f'past_thr={x[x > thr].mean():.3f}-mean: ', end='')
s = f'{PREFIX}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr\n' \
f'{PREFIX}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, ' \
f'past_thr={x[x > thr].mean():.3f}-mean: '
for i, x in enumerate(k):
print('%i,%i' % (round(x[0]), round(x[1])), end=', ' if i < len(k) - 1 else '\n') # use in *.cfg
s += '%i,%i, ' % (round(x[0]), round(x[1]))
if verbose:
LOGGER.info(s[:-2])
return k
if isinstance(path, str): # *.yaml file
with open(path) as f:
if isinstance(dataset, str): # *.yaml file
with open(dataset, errors='ignore') as f:
data_dict = yaml.safe_load(f) # model dict
from utils.datasets import LoadImagesAndLabels
dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
else:
dataset = path # dataset
# Get label wh
shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True)
@@ -116,19 +119,19 @@ def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=10
# Filter
i = (wh0 < 3.0).any(1).sum()
if i:
print(f'{prefix}WARNING: Extremely small objects found. {i} of {len(wh0)} labels are < 3 pixels in size.')
LOGGER.info(f'{PREFIX}WARNING: Extremely small objects found. {i} of {len(wh0)} labels are < 3 pixels in size.')
wh = wh0[(wh0 >= 2.0).any(1)] # filter > 2 pixels
# wh = wh * (np.random.rand(wh.shape[0], 1) * 0.9 + 0.1) # multiply by random scale 0-1
# Kmeans calculation
print(f'{prefix}Running kmeans for {n} anchors on {len(wh)} points...')
LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(wh)} points...')
s = wh.std(0) # sigmas for whitening
k, dist = kmeans(wh / s, n, iter=30) # points, mean distance
assert len(k) == n, print(f'{prefix}ERROR: scipy.cluster.vq.kmeans requested {n} points but returned only {len(k)}')
assert len(k) == n, f'{PREFIX}ERROR: scipy.cluster.vq.kmeans requested {n} points but returned only {len(k)}'
k *= s
wh = torch.tensor(wh, dtype=torch.float32) # filtered
wh0 = torch.tensor(wh0, dtype=torch.float32) # unfiltered
k = print_results(k)
k = print_results(k, verbose=False)
# Plot
# k, d = [None] * 20, [None] * 20
@@ -145,17 +148,17 @@ def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=10
# Evolve
npr = np.random
f, sh, mp, s = anchor_fitness(k), k.shape, 0.9, 0.1 # fitness, generations, mutation prob, sigma
pbar = tqdm(range(gen), desc=f'{prefix}Evolving anchors with Genetic Algorithm:') # progress bar
pbar = tqdm(range(gen), desc=f'{PREFIX}Evolving anchors with Genetic Algorithm:') # progress bar
for _ in pbar:
v = np.ones(sh)
while (v == 1).all(): # mutate until a change occurs (prevent duplicates)
v = ((npr.random(sh) < mp) * npr.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
v = ((npr.random(sh) < mp) * random.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
kg = (k.copy() * v).clip(min=2.0)
fg = anchor_fitness(kg)
if fg > f:
f, k = fg, kg.copy()
pbar.desc = f'{prefix}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}'
pbar.desc = f'{PREFIX}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}'
if verbose:
print_results(k)
print_results(k, verbose)
return print_results(k)
+57
View File
@@ -0,0 +1,57 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Auto-batch utils
"""
from copy import deepcopy
import numpy as np
import torch
from torch.cuda import amp
from utils.general import LOGGER, colorstr
from utils.torch_utils import profile
def check_train_batch_size(model, imgsz=640):
# Check training batch size
with amp.autocast():
return autobatch(deepcopy(model).train(), imgsz) # compute optimal batch size
def autobatch(model, imgsz=640, fraction=0.9, batch_size=16):
# Automatically estimate best batch size to use `fraction` of available CUDA memory
# Usage:
# import torch
# from utils.autobatch import autobatch
# model = torch.hub.load('ultralytics/yolov3', 'yolov3', autoshape=False)
# print(autobatch(model))
prefix = colorstr('AutoBatch: ')
LOGGER.info(f'{prefix}Computing optimal batch size for --imgsz {imgsz}')
device = next(model.parameters()).device # get model device
if device.type == 'cpu':
LOGGER.info(f'{prefix}CUDA not detected, using default CPU batch-size {batch_size}')
return batch_size
d = str(device).upper() # 'CUDA:0'
properties = torch.cuda.get_device_properties(device) # device properties
t = properties.total_memory / 1024 ** 3 # (GiB)
r = torch.cuda.memory_reserved(device) / 1024 ** 3 # (GiB)
a = torch.cuda.memory_allocated(device) / 1024 ** 3 # (GiB)
f = t - (r + a) # free inside reserved
LOGGER.info(f'{prefix}{d} ({properties.name}) {t:.2f}G total, {r:.2f}G reserved, {a:.2f}G allocated, {f:.2f}G free')
batch_sizes = [1, 2, 4, 8, 16]
try:
img = [torch.zeros(b, 3, imgsz, imgsz) for b in batch_sizes]
y = profile(img, model, n=3, device=device)
except Exception as e:
LOGGER.warning(f'{prefix}{e}')
y = [x[2] for x in y if x] # memory [2]
batch_sizes = batch_sizes[:len(y)]
p = np.polyfit(batch_sizes, y, deg=1) # first degree polynomial fit
b = int((f * fraction - p[1]) / p[0]) # y intercept (optimal batch size)
LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%)')
return b
-26
View File
@@ -1,26 +0,0 @@
# AWS EC2 instance startup 'MIME' script https://aws.amazon.com/premiumsupport/knowledge-center/execute-user-data-ec2/
# This script will run on every instance restart, not only on first start
# --- DO NOT COPY ABOVE COMMENTS WHEN PASTING INTO USERDATA ---
Content-Type: multipart/mixed; boundary="//"
MIME-Version: 1.0
--//
Content-Type: text/cloud-config; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="cloud-config.txt"
#cloud-config
cloud_final_modules:
- [scripts-user, always]
--//
Content-Type: text/x-shellscript; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="userdata.txt"
#!/bin/bash
# --- paste contents of userdata.sh here ---
--//
-37
View File
@@ -1,37 +0,0 @@
# Resume all interrupted trainings in yolov5/ dir including DDP trainings
# Usage: $ python utils/aws/resume.py
import os
import sys
from pathlib import Path
import torch
import yaml
sys.path.append('./') # to run '$ python *.py' files in subdirectories
port = 0 # --master_port
path = Path('').resolve()
for last in path.rglob('*/**/last.pt'):
ckpt = torch.load(last)
if ckpt['optimizer'] is None:
continue
# Load opt.yaml
with open(last.parent.parent / 'opt.yaml') as f:
opt = yaml.safe_load(f)
# Get device count
d = opt['device'].split(',') # devices
nd = len(d) # number of devices
ddp = nd > 1 or (nd == 0 and torch.cuda.device_count() > 1) # distributed data parallel
if ddp: # multi-GPU
port += 1
cmd = f'python -m torch.distributed.launch --nproc_per_node {nd} --master_port {port} train.py --resume {last}'
else: # single-GPU
cmd = f'python train.py --resume {last}'
cmd += ' > /dev/null 2>&1 &' # redirect output to dev/null and run in daemon thread
print(cmd)
os.system(cmd)
-27
View File
@@ -1,27 +0,0 @@
#!/bin/bash
# AWS EC2 instance startup script https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html
# This script will run only once on first instance start (for a re-start script see mime.sh)
# /home/ubuntu (ubuntu) or /home/ec2-user (amazon-linux) is working dir
# Use >300 GB SSD
cd home/ubuntu
if [ ! -d yolov5 ]; then
echo "Running first-time script." # install dependencies, download COCO, pull Docker
git clone https://github.com/ultralytics/yolov5 -b master && sudo chmod -R 777 yolov5
cd yolov5
bash data/scripts/get_coco.sh && echo "Data done." &
sudo docker pull ultralytics/yolov5:latest && echo "Docker done." &
python -m pip install --upgrade pip && pip install -r requirements.txt && python detect.py && echo "Requirements done." &
wait && echo "All tasks done." # finish background tasks
else
echo "Running re-start script." # resume interrupted runs
i=0
list=$(sudo docker ps -qa) # container list i.e. $'one\ntwo\nthree\nfour'
while IFS= read -r id; do
((i++))
echo "restarting container $i: $id"
sudo docker start $id
# sudo docker exec -it $id python train.py --resume # single-GPU
sudo docker exec -d $id python utils/aws/resume.py # multi-scenario
done <<<"$list"
fi
+76
View File
@@ -0,0 +1,76 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Callback utils
"""
class Callbacks:
""""
Handles all registered callbacks for Hooks
"""
# Define the available callbacks
_callbacks = {
'on_pretrain_routine_start': [],
'on_pretrain_routine_end': [],
'on_train_start': [],
'on_train_epoch_start': [],
'on_train_batch_start': [],
'optimizer_step': [],
'on_before_zero_grad': [],
'on_train_batch_end': [],
'on_train_epoch_end': [],
'on_val_start': [],
'on_val_batch_start': [],
'on_val_image_end': [],
'on_val_batch_end': [],
'on_val_end': [],
'on_fit_epoch_end': [], # fit = train + val
'on_model_save': [],
'on_train_end': [],
'teardown': [],
}
def register_action(self, hook, name='', callback=None):
"""
Register a new action to a callback hook
Args:
hook The callback hook name to register the action to
name The name of the action for later reference
callback The callback to fire
"""
assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
assert callable(callback), f"callback '{callback}' is not callable"
self._callbacks[hook].append({'name': name, 'callback': callback})
def get_registered_actions(self, hook=None):
""""
Returns all the registered actions by callback hook
Args:
hook The name of the hook to check, defaults to all
"""
if hook:
return self._callbacks[hook]
else:
return self._callbacks
def run(self, hook, *args, **kwargs):
"""
Loop through the registered actions and fire all callbacks
Args:
hook The name of the hook to check, defaults to all
args Arguments to receive from
kwargs Keyword Arguments to receive from
"""
assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
for logger in self._callbacks[hook]:
logger['callback'](*args, **kwargs)
+406 -441
View File
File diff suppressed because it is too large Load Diff
+22 -12
View File
@@ -1,10 +1,15 @@
# Google utils: https://cloud.google.com/storage/docs/reference/libraries
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Download utils
"""
import os
import platform
import subprocess
import time
import urllib
from pathlib import Path
from zipfile import ZipFile
import requests
import torch
@@ -19,30 +24,32 @@ def gsutil_getsize(url=''):
def safe_download(file, url, url2=None, min_bytes=1E0, error_msg=''):
# Attempts to download file from url or url2, checks and removes incomplete downloads < min_bytes
file = Path(file)
try: # GitHub
assert_msg = f"Downloaded file '{file}' does not exist or size is < min_bytes={min_bytes}"
try: # url1
print(f'Downloading {url} to {file}...')
torch.hub.download_url_to_file(url, str(file))
assert file.exists() and file.stat().st_size > min_bytes # check
except Exception as e: # GCP
assert file.exists() and file.stat().st_size > min_bytes, assert_msg # check
except Exception as e: # url2
file.unlink(missing_ok=True) # remove partial downloads
print(f'Download error: {e}\nRe-attempting {url2 or url} to {file}...')
print(f'ERROR: {e}\nRe-attempting {url2 or url} to {file}...')
os.system(f"curl -L '{url2 or url}' -o '{file}' --retry 3 -C -") # curl download, retry and resume on fail
finally:
if not file.exists() or file.stat().st_size < min_bytes: # check
file.unlink(missing_ok=True) # remove partial downloads
print(f'ERROR: Download failure: {error_msg or url}')
print(f"ERROR: {assert_msg}\n{error_msg}")
print('')
def attempt_download(file, repo='ultralytics/yolov3'):
def attempt_download(file, repo='ultralytics/yolov3'): # from utils.downloads import *; attempt_download()
# Attempt file download if does not exist
file = Path(str(file).strip().replace("'", ''))
if not file.exists():
# URL specified
name = file.name
name = Path(urllib.parse.unquote(str(file))).name # decode '%2F' to '/' etc.
if str(file).startswith(('http:/', 'https:/')): # download
url = str(file).replace(':/', '://') # Pathlib turns :// -> :/
name = name.split('?')[0] # parse authentication https://url.com/file.txt?auth...
safe_download(file=name, url=url, min_bytes=1E5)
return name
@@ -50,7 +57,7 @@ def attempt_download(file, repo='ultralytics/yolov3'):
file.parent.mkdir(parents=True, exist_ok=True) # make parent dir (if required)
try:
response = requests.get(f'https://api.github.com/repos/{repo}/releases/latest').json() # github api
assets = [x['name'] for x in response['assets']] # release assets, i.e. ['yolov5s.pt', 'yolov5m.pt', ...]
assets = [x['name'] for x in response['assets']] # release assets, i.e. ['yolov3.pt'...]
tag = response['tag_name'] # i.e. 'v1.0'
except: # fallback plan
assets = ['yolov3.pt', 'yolov3-spp.pt', 'yolov3-tiny.pt']
@@ -70,7 +77,7 @@ def attempt_download(file, repo='ultralytics/yolov3'):
def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'):
# Downloads a file from Google Drive. from yolov3.utils.google_utils import *; gdrive_download()
# Downloads a file from Google Drive. from yolov3.utils.downloads import *; gdrive_download()
t = time.time()
file = Path(file)
cookie = Path('cookie') # gdrive cookie
@@ -97,8 +104,8 @@ def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'):
# Unzip if archive
if file.suffix == '.zip':
print('unzipping... ', end='')
os.system(f'unzip -q {file}') # unzip
file.unlink() # remove zip to free space
ZipFile(file).extractall(path=file.parent) # unzip
file.unlink() # remove zip
print(f'Done ({time.time() - t:.1f}s)')
return r
@@ -111,6 +118,9 @@ def get_token(cookie="./cookie"):
return line.split()[-1]
return ""
# Google utils: https://cloud.google.com/storage/docs/reference/libraries ----------------------------------------------
#
#
# def upload_blob(bucket_name, source_file_name, destination_blob_name):
# # Uploads a file to a bucket
# # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python
-68
View File
@@ -1,68 +0,0 @@
# Flask REST API
[REST](https://en.wikipedia.org/wiki/Representational_state_transfer) [API](https://en.wikipedia.org/wiki/API)s are commonly used to expose Machine Learning (ML) models to other services. This folder contains an example REST API created using Flask to expose the YOLOv5s model from [PyTorch Hub](https://pytorch.org/hub/ultralytics_yolov5/).
## Requirements
[Flask](https://palletsprojects.com/p/flask/) is required. Install with:
```shell
$ pip install Flask
```
## Run
After Flask installation run:
```shell
$ python3 restapi.py --port 5000
```
Then use [curl](https://curl.se/) to perform a request:
```shell
$ curl -X POST -F image=@zidane.jpg 'http://localhost:5000/v1/object-detection/yolov5s'`
```
The model inference results are returned as a JSON response:
```json
[
{
"class": 0,
"confidence": 0.8900438547,
"height": 0.9318675399,
"name": "person",
"width": 0.3264600933,
"xcenter": 0.7438579798,
"ycenter": 0.5207948685
},
{
"class": 0,
"confidence": 0.8440024257,
"height": 0.7155083418,
"name": "person",
"width": 0.6546785235,
"xcenter": 0.427829951,
"ycenter": 0.6334488392
},
{
"class": 27,
"confidence": 0.3771208823,
"height": 0.3902671337,
"name": "tie",
"width": 0.0696444362,
"xcenter": 0.3675483763,
"ycenter": 0.7991207838
},
{
"class": 27,
"confidence": 0.3527112305,
"height": 0.1540903747,
"name": "tie",
"width": 0.0336618312,
"xcenter": 0.7814827561,
"ycenter": 0.5065554976
}
]
```
An example python script to perform inference using [requests](https://docs.python-requests.org/en/master/) is given in `example_request.py`
-13
View File
@@ -1,13 +0,0 @@
"""Perform test request"""
import pprint
import requests
DETECTION_URL = "http://localhost:5000/v1/object-detection/yolov5s"
TEST_IMAGE = "zidane.jpg"
image_data = open(TEST_IMAGE, "rb").read()
response = requests.post(DETECTION_URL, files={"image": image_data}).json()
pprint.pprint(response)
-37
View File
@@ -1,37 +0,0 @@
"""
Run a rest API exposing the yolov5s object detection model
"""
import argparse
import io
import torch
from PIL import Image
from flask import Flask, request
app = Flask(__name__)
DETECTION_URL = "/v1/object-detection/yolov5s"
@app.route(DETECTION_URL, methods=["POST"])
def predict():
if not request.method == "POST":
return
if request.files.get("image"):
image_file = request.files["image"]
image_bytes = image_file.read()
img = Image.open(io.BytesIO(image_bytes))
results = model(img, size=640) # reduce size=320 for faster inference
return results.pandas().xyxy[0].to_json(orient="records")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Flask API exposing YOLOv3 model")
parser.add_argument("--port", default=5000, type=int, help="port number")
args = parser.parse_args()
model = torch.hub.load("ultralytics/yolov5", "yolov5s", force_reload=True) # force_reload to recache
app.run(host="0.0.0.0", port=args.port) # debug=True causes Restarting with stat
+352 -210
View File
@@ -1,5 +1,9 @@
# YOLOv3 general utils
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
General utils
"""
import contextlib
import glob
import logging
import math
@@ -7,11 +11,15 @@ import os
import platform
import random
import re
import subprocess
import shutil
import signal
import time
import urllib
from itertools import repeat
from multiprocessing.pool import ThreadPool
from pathlib import Path
from subprocess import check_output
from zipfile import ZipFile
import cv2
import numpy as np
@@ -21,9 +29,8 @@ import torch
import torchvision
import yaml
from utils.google_utils import gsutil_getsize
from utils.metrics import fitness
from utils.torch_utils import init_torch_seeds
from utils.downloads import gsutil_getsize
from utils.metrics import box_iou, fitness
# Settings
torch.set_printoptions(linewidth=320, precision=5, profile='long')
@@ -32,18 +39,96 @@ pd.options.display.max_columns = 10
cv2.setNumThreads(0) # prevent OpenCV from multithreading (incompatible with PyTorch DataLoader)
os.environ['NUMEXPR_MAX_THREADS'] = str(min(os.cpu_count(), 8)) # NumExpr max threads
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # root directory
def set_logging(rank=-1, verbose=True):
logging.basicConfig(
format="%(message)s",
level=logging.INFO if (verbose and rank in [-1, 0]) else logging.WARN)
def set_logging(name=None, verbose=True):
# Sets level and returns logger
rank = int(os.getenv('RANK', -1)) # rank in world for Multi-GPU trainings
logging.basicConfig(format="%(message)s", level=logging.INFO if (verbose and rank in (-1, 0)) else logging.WARNING)
return logging.getLogger(name)
LOGGER = set_logging(__name__) # define globally (used in train.py, val.py, detect.py, etc.)
class Profile(contextlib.ContextDecorator):
# Usage: @Profile() decorator or 'with Profile():' context manager
def __enter__(self):
self.start = time.time()
def __exit__(self, type, value, traceback):
print(f'Profile results: {time.time() - self.start:.5f}s')
class Timeout(contextlib.ContextDecorator):
# Usage: @Timeout(seconds) decorator or 'with Timeout(seconds):' context manager
def __init__(self, seconds, *, timeout_msg='', suppress_timeout_errors=True):
self.seconds = int(seconds)
self.timeout_message = timeout_msg
self.suppress = bool(suppress_timeout_errors)
def _timeout_handler(self, signum, frame):
raise TimeoutError(self.timeout_message)
def __enter__(self):
signal.signal(signal.SIGALRM, self._timeout_handler) # Set handler for SIGALRM
signal.alarm(self.seconds) # start countdown for SIGALRM to be raised
def __exit__(self, exc_type, exc_val, exc_tb):
signal.alarm(0) # Cancel SIGALRM if it's scheduled
if self.suppress and exc_type is TimeoutError: # Suppress TimeoutError
return True
class WorkingDirectory(contextlib.ContextDecorator):
# Usage: @WorkingDirectory(dir) decorator or 'with WorkingDirectory(dir):' context manager
def __init__(self, new_dir):
self.dir = new_dir # new dir
self.cwd = Path.cwd().resolve() # current dir
def __enter__(self):
os.chdir(self.dir)
def __exit__(self, exc_type, exc_val, exc_tb):
os.chdir(self.cwd)
def try_except(func):
# try-except function. Usage: @try_except decorator
def handler(*args, **kwargs):
try:
func(*args, **kwargs)
except Exception as e:
print(e)
return handler
def methods(instance):
# Get class/instance methods
return [f for f in dir(instance) if callable(getattr(instance, f)) and not f.startswith("__")]
def print_args(name, opt):
# Print argparser arguments
LOGGER.info(colorstr(f'{name}: ') + ', '.join(f'{k}={v}' for k, v in vars(opt).items()))
def init_seeds(seed=0):
# Initialize random number generator (RNG) seeds
# Initialize random number generator (RNG) seeds https://pytorch.org/docs/stable/notes/randomness.html
# cudnn seed 0 settings are slower and more reproducible, else faster and less reproducible
import torch.backends.cudnn as cudnn
random.seed(seed)
np.random.seed(seed)
init_torch_seeds(seed)
torch.manual_seed(seed)
cudnn.benchmark, cudnn.deterministic = (False, True) if seed == 0 else (True, False)
def intersect_dicts(da, db, exclude=()):
# Dictionary intersection of matching keys and shapes, omitting 'exclude' keys, using da values
return {k: v for k, v in da.items() if k in db and not any(x in k for x in exclude) and v.shape == db[k].shape}
def get_latest_run(search_dir='.'):
@@ -52,81 +137,136 @@ def get_latest_run(search_dir='.'):
return max(last_list, key=os.path.getctime) if last_list else ''
def user_config_dir(dir='Ultralytics', env_var='YOLOV3_CONFIG_DIR'):
# Return path of user configuration directory. Prefer environment variable if exists. Make dir if required.
env = os.getenv(env_var)
if env:
path = Path(env) # use environment variable
else:
cfg = {'Windows': 'AppData/Roaming', 'Linux': '.config', 'Darwin': 'Library/Application Support'} # 3 OS dirs
path = Path.home() / cfg.get(platform.system(), '') # OS-specific config dir
path = (path if is_writeable(path) else Path('/tmp')) / dir # GCP and AWS lambda fix, only /tmp is writeable
path.mkdir(exist_ok=True) # make if required
return path
def is_writeable(dir, test=False):
# Return True if directory has write permissions, test opening a file with write permissions if test=True
if test: # method 1
file = Path(dir) / 'tmp.txt'
try:
with open(file, 'w'): # open file with write permissions
pass
file.unlink() # remove file
return True
except OSError:
return False
else: # method 2
return os.access(dir, os.R_OK) # possible issues on Windows
def is_docker():
# Is environment a Docker container
# Is environment a Docker container?
return Path('/workspace').exists() # or Path('/.dockerenv').exists()
def is_colab():
# Is environment a Google Colab instance
# Is environment a Google Colab instance?
try:
import google.colab
return True
except Exception as e:
except ImportError:
return False
def is_pip():
# Is file in a pip package?
return 'site-packages' in Path(__file__).resolve().parts
def is_ascii(s=''):
# Is string composed of all ASCII (no UTF) characters? (note str().isascii() introduced in python 3.7)
s = str(s) # convert list, tuple, None, etc. to str
return len(s.encode().decode('ascii', 'ignore')) == len(s)
def is_chinese(s='人工智能'):
# Is string composed of any Chinese characters?
return re.search('[\u4e00-\u9fff]', s)
def emojis(str=''):
# Return platform-dependent emoji-safe version of string
return str.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else str
def file_size(file):
# Return file size in MB
return Path(file).stat().st_size / 1e6
def file_size(path):
# Return file/dir size (MB)
path = Path(path)
if path.is_file():
return path.stat().st_size / 1E6
elif path.is_dir():
return sum(f.stat().st_size for f in path.glob('**/*') if f.is_file()) / 1E6
else:
return 0.0
def check_online():
# Check internet connectivity
import socket
try:
socket.create_connection(("1.1.1.1", 443), 5) # check host accesability
socket.create_connection(("1.1.1.1", 443), 5) # check host accessibility
return True
except OSError:
return False
@try_except
@WorkingDirectory(ROOT)
def check_git_status():
# Recommend 'git pull' if code is out of date
msg = ', for updates see https://github.com/ultralytics/yolov3'
print(colorstr('github: '), end='')
try:
assert Path('.git').exists(), 'skipping check (not a git repository)'
assert not is_docker(), 'skipping check (Docker image)'
assert check_online(), 'skipping check (offline)'
assert Path('.git').exists(), 'skipping check (not a git repository)' + msg
assert not is_docker(), 'skipping check (Docker image)' + msg
assert check_online(), 'skipping check (offline)' + msg
cmd = 'git fetch && git config --get remote.origin.url'
url = subprocess.check_output(cmd, shell=True).decode().strip().rstrip('.git') # github repo url
branch = subprocess.check_output('git rev-parse --abbrev-ref HEAD', shell=True).decode().strip() # checked out
n = int(subprocess.check_output(f'git rev-list {branch}..origin/master --count', shell=True)) # commits behind
if n > 0:
s = f"⚠️ WARNING: code is out of date by {n} commit{'s' * (n > 1)}. " \
f"Use 'git pull' to update or 'git clone {url}' to download latest."
else:
s = f'up to date with {url}'
print(emojis(s)) # emoji-safe
except Exception as e:
print(e)
cmd = 'git fetch && git config --get remote.origin.url'
url = check_output(cmd, shell=True, timeout=5).decode().strip().rstrip('.git') # git fetch
branch = check_output('git rev-parse --abbrev-ref HEAD', shell=True).decode().strip() # checked out
n = int(check_output(f'git rev-list {branch}..origin/master --count', shell=True)) # commits behind
if n > 0:
s = f"⚠️ YOLOv3 is out of date by {n} commit{'s' * (n > 1)}. Use `git pull` or `git clone {url}` to update."
else:
s = f'up to date with {url}'
print(emojis(s)) # emoji-safe
def check_python(minimum='3.7.0', required=True):
def check_python(minimum='3.6.2'):
# Check current python version vs. required python version
current = platform.python_version()
result = pkg.parse_version(current) >= pkg.parse_version(minimum)
if required:
assert result, f'Python {minimum} required by YOLOv3, but Python {current} is currently installed'
return result
check_version(platform.python_version(), minimum, name='Python ', hard=True)
def check_requirements(requirements='requirements.txt', exclude=()):
def check_version(current='0.0.0', minimum='0.0.0', name='version ', pinned=False, hard=False):
# Check version vs. required version
current, minimum = (pkg.parse_version(x) for x in (current, minimum))
result = (current == minimum) if pinned else (current >= minimum) # bool
if hard: # assert min requirements met
assert result, f'{name}{minimum} required by YOLOv3, but {name}{current} is currently installed'
else:
return result
@try_except
def check_requirements(requirements=ROOT / 'requirements.txt', exclude=(), install=True):
# Check installed dependencies meet requirements (pass *.txt file or list of packages)
prefix = colorstr('red', 'bold', 'requirements:')
check_python() # check python version
if isinstance(requirements, (str, Path)): # requirements.txt file
file = Path(requirements)
if not file.exists():
print(f"{prefix} {file.resolve()} not found, check failed.")
return
requirements = [f'{x.name}{x.specifier}' for x in pkg.parse_requirements(file.open()) if x.name not in exclude]
assert file.exists(), f"{prefix} {file.resolve()} not found, check failed."
with file.open() as f:
requirements = [f'{x.name}{x.specifier}' for x in pkg.parse_requirements(f) if x.name not in exclude]
else: # list or tuple of packages
requirements = [x for x in requirements if x not in exclude]
@@ -135,25 +275,33 @@ def check_requirements(requirements='requirements.txt', exclude=()):
try:
pkg.require(r)
except Exception as e: # DistributionNotFound or VersionConflict if requirements not met
n += 1
print(f"{prefix} {r} not found and is required by YOLOv3, attempting auto-update...")
try:
print(subprocess.check_output(f"pip install '{r}'", shell=True).decode())
except Exception as e:
print(f'{prefix} {e}')
s = f"{prefix} {r} not found and is required by YOLOv3"
if install:
print(f"{s}, attempting auto-update...")
try:
assert check_online(), f"'pip install {r}' skipped (offline)"
print(check_output(f"pip install '{r}'", shell=True).decode())
n += 1
except Exception as e:
print(f'{prefix} {e}')
else:
print(f'{s}. Please install and rerun your command.')
if n: # if packages updated
source = file.resolve() if 'file' in locals() else requirements
s = f"{prefix} {n} package{'s' * (n > 1)} updated per {source}\n" \
f"{prefix} ⚠️ {colorstr('bold', 'Restart runtime or rerun command for updates to take effect')}\n"
print(emojis(s)) # emoji-safe
print(emojis(s))
def check_img_size(img_size, s=32):
# Verify img_size is a multiple of stride s
new_size = make_divisible(img_size, int(s)) # ceil gs-multiple
if new_size != img_size:
print('WARNING: --img-size %g must be multiple of max stride %g, updating to %g' % (img_size, s, new_size))
def check_img_size(imgsz, s=32, floor=0):
# Verify image size is a multiple of stride s in each dimension
if isinstance(imgsz, int): # integer i.e. img_size=640
new_size = max(make_divisible(imgsz, int(s)), floor)
else: # list i.e. img_size=[640, 480]
new_size = [max(make_divisible(x, int(s)), floor) for x in imgsz]
if new_size != imgsz:
print(f'WARNING: --img-size {imgsz} must be multiple of max stride {s}, updating to {new_size}')
return new_size
@@ -172,53 +320,114 @@ def check_imshow():
return False
def check_file(file):
def check_suffix(file='yolov3.pt', suffix=('.pt',), msg=''):
# Check file(s) for acceptable suffix
if file and suffix:
if isinstance(suffix, str):
suffix = [suffix]
for f in file if isinstance(file, (list, tuple)) else [file]:
s = Path(f).suffix.lower() # file suffix
if len(s):
assert s in suffix, f"{msg}{f} acceptable suffix is {suffix}"
def check_yaml(file, suffix=('.yaml', '.yml')):
# Search/download YAML file (if necessary) and return path, checking suffix
return check_file(file, suffix)
def check_file(file, suffix=''):
# Search/download file (if necessary) and return path
check_suffix(file, suffix) # optional
file = str(file) # convert to str()
if Path(file).is_file() or file == '': # exists
return file
elif file.startswith(('http://', 'https://')): # download
url, file = file, Path(file).name
print(f'Downloading {url} to {file}...')
torch.hub.download_url_to_file(url, file)
assert Path(file).exists() and Path(file).stat().st_size > 0, f'File download failed: {url}' # check
elif file.startswith(('http:/', 'https:/')): # download
url = str(Path(file)).replace(':/', '://') # Pathlib turns :// -> :/
file = Path(urllib.parse.unquote(file).split('?')[0]).name # '%2F' to '/', split https://url.com/file.txt?auth
if Path(file).is_file():
print(f'Found {url} locally at {file}') # file already exists
else:
print(f'Downloading {url} to {file}...')
torch.hub.download_url_to_file(url, file)
assert Path(file).exists() and Path(file).stat().st_size > 0, f'File download failed: {url}' # check
return file
else: # search
files = glob.glob('./**/' + file, recursive=True) # find file
files = []
for d in 'data', 'models', 'utils': # search directories
files.extend(glob.glob(str(ROOT / d / '**' / file), recursive=True)) # find file
assert len(files), f'File not found: {file}' # assert file was found
assert len(files) == 1, f"Multiple files match '{file}', specify exact path: {files}" # assert unique
return files[0] # return file
def check_dataset(dict):
# Download dataset if not found locally
val, s = dict.get('val'), dict.get('download')
if val and len(val):
def check_dataset(data, autodownload=True):
# Download and/or unzip dataset if not found locally
# Usage: https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128_with_yaml.zip
# Download (optional)
extract_dir = ''
if isinstance(data, (str, Path)) and str(data).endswith('.zip'): # i.e. gs://bucket/dir/coco128.zip
download(data, dir='../datasets', unzip=True, delete=False, curl=False, threads=1)
data = next((Path('../datasets') / Path(data).stem).rglob('*.yaml'))
extract_dir, autodownload = data.parent, False
# Read yaml (optional)
if isinstance(data, (str, Path)):
with open(data, errors='ignore') as f:
data = yaml.safe_load(f) # dictionary
# Parse yaml
path = extract_dir or Path(data.get('path') or '') # optional 'path' default to '.'
for k in 'train', 'val', 'test':
if data.get(k): # prepend path
data[k] = str(path / data[k]) if isinstance(data[k], str) else [str(path / x) for x in data[k]]
assert 'nc' in data, "Dataset 'nc' key missing."
if 'names' not in data:
data['names'] = [f'class{i}' for i in range(data['nc'])] # assign class names if missing
train, val, test, s = (data.get(x) for x in ('train', 'val', 'test', 'download'))
if val:
val = [Path(x).resolve() for x in (val if isinstance(val, list) else [val])] # val path
if not all(x.exists() for x in val):
print('\nWARNING: Dataset not found, nonexistent paths: %s' % [str(x) for x in val if not x.exists()])
if s and len(s): # download script
if s and autodownload: # download script
root = path.parent if 'path' in data else '..' # unzip directory i.e. '../'
if s.startswith('http') and s.endswith('.zip'): # URL
f = Path(s).name # filename
print(f'Downloading {s} ...')
print(f'Downloading {s} to {f}...')
torch.hub.download_url_to_file(s, f)
r = os.system(f'unzip -q {f} -d ../ && rm {f}') # unzip
Path(root).mkdir(parents=True, exist_ok=True) # create root
ZipFile(f).extractall(path=root) # unzip
Path(f).unlink() # remove zip
r = None # success
elif s.startswith('bash '): # bash script
print(f'Running {s} ...')
r = os.system(s)
else: # python script
r = exec(s) # return None
print('Dataset autodownload %s\n' % ('success' if r in (0, None) else 'failure')) # print result
r = exec(s, {'yaml': data}) # return None
print(f"Dataset autodownload {f'success, saved to {root}' if r in (0, None) else 'failure'}\n")
else:
raise Exception('Dataset not found.')
return data # dictionary
def url2file(url):
# Convert URL to filename, i.e. https://url.com/file.txt?auth -> file.txt
url = str(Path(url)).replace(':/', '://') # Pathlib turns :// -> :/
file = Path(urllib.parse.unquote(url)).name.split('?')[0] # '%2F' to '/', split https://url.com/file.txt?auth
return file
def download(url, dir='.', unzip=True, delete=True, curl=False, threads=1):
# Multi-threaded file download and unzip function
# Multi-threaded file download and unzip function, used in data.yaml for autodownload
def download_one(url, dir):
# Download 1 file
f = dir / Path(url).name # filename
if not f.exists():
if Path(url).is_file(): # exists in current path
Path(url).rename(f) # move to dir
elif not f.exists():
print(f'Downloading {url} to {f}...')
if curl:
os.system(f"curl -L '{url}' -o '{f}' --retry 9 -C -") # curl download, retry and resume on fail
@@ -227,12 +436,11 @@ def download(url, dir='.', unzip=True, delete=True, curl=False, threads=1):
if unzip and f.suffix in ('.zip', '.gz'):
print(f'Unzipping {f}...')
if f.suffix == '.zip':
s = f'unzip -qo {f} -d {dir} && rm {f}' # unzip -quiet -overwrite
ZipFile(f).extractall(path=dir) # unzip
elif f.suffix == '.gz':
s = f'tar xfz {f} --directory {f.parent}' # unzip
if delete: # delete zip file after unzip
s += f' && rm {f}'
os.system(s)
os.system(f'tar xfz {f} --directory {f.parent}') # unzip
if delete:
f.unlink() # remove zip
dir = Path(dir)
dir.mkdir(parents=True, exist_ok=True) # make directory
@@ -242,7 +450,7 @@ def download(url, dir='.', unzip=True, delete=True, curl=False, threads=1):
pool.close()
pool.join()
else:
for u in tuple(url) if isinstance(url, str) else url:
for u in [url] if isinstance(url, (str, Path)) else url:
download_one(u, dir)
@@ -257,7 +465,7 @@ def clean_str(s):
def one_cycle(y1=0.0, y2=1.0, steps=100):
# lambda function for sinusoidal ramp from y1 to y2
# lambda function for sinusoidal ramp from y1 to y2 https://arxiv.org/pdf/1812.01187.pdf
return lambda x: ((1 - math.cos(x * math.pi / steps)) / 2) * (y2 - y1) + y1
@@ -355,6 +563,18 @@ def xywhn2xyxy(x, w=640, h=640, padw=0, padh=0):
return y
def xyxy2xywhn(x, w=640, h=640, clip=False, eps=0.0):
# Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] normalized where xy1=top-left, xy2=bottom-right
if clip:
clip_coords(x, (h - eps, w - eps)) # warning: inplace clip
y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
y[:, 0] = ((x[:, 0] + x[:, 2]) / 2) / w # x center
y[:, 1] = ((x[:, 1] + x[:, 3]) / 2) / h # y center
y[:, 2] = (x[:, 2] - x[:, 0]) / w # width
y[:, 3] = (x[:, 3] - x[:, 1]) / h # height
return y
def xyn2xy(x, w=640, h=640, padw=0, padh=0):
# Convert normalized segments into pixel segments, shape (n,2)
y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
@@ -405,90 +625,16 @@ def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
return coords
def clip_coords(boxes, img_shape):
def clip_coords(boxes, shape):
# Clip bounding xyxy bounding boxes to image shape (height, width)
boxes[:, 0].clamp_(0, img_shape[1]) # x1
boxes[:, 1].clamp_(0, img_shape[0]) # y1
boxes[:, 2].clamp_(0, img_shape[1]) # x2
boxes[:, 3].clamp_(0, img_shape[0]) # y2
def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7):
# Returns the IoU of box1 to box2. box1 is 4, box2 is nx4
box2 = box2.T
# Get the coordinates of bounding boxes
if x1y1x2y2: # x1, y1, x2, y2 = box1
b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
else: # transform from xywh to xyxy
b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2
b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2
b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2
b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2
# Intersection area
inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
(torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
# Union Area
w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps
union = w1 * h1 + w2 * h2 - inter + eps
iou = inter / union
if GIoU or DIoU or CIoU:
cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex (smallest enclosing box) width
ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height
if CIoU or DIoU: # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
c2 = cw ** 2 + ch ** 2 + eps # convex diagonal squared
rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 +
(b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4 # center distance squared
if DIoU:
return iou - rho2 / c2 # DIoU
elif CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
with torch.no_grad():
alpha = v / (v - iou + (1 + eps))
return iou - (rho2 / c2 + v * alpha) # CIoU
else: # GIoU https://arxiv.org/pdf/1902.09630.pdf
c_area = cw * ch + eps # convex area
return iou - (c_area - union) / c_area # GIoU
else:
return iou # IoU
def box_iou(box1, box2):
# https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
"""
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
box1 (Tensor[N, 4])
box2 (Tensor[M, 4])
Returns:
iou (Tensor[N, M]): the NxM matrix containing the pairwise
IoU values for every element in boxes1 and boxes2
"""
def box_area(box):
# box = 4xn
return (box[2] - box[0]) * (box[3] - box[1])
area1 = box_area(box1.T)
area2 = box_area(box2.T)
# inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
return inter / (area1[:, None] + area2 - inter) # iou = inter / (area1 + area2 - inter)
def wh_iou(wh1, wh2):
# Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
wh1 = wh1[:, None] # [N,1,2]
wh2 = wh2[None] # [1,M,2]
inter = torch.min(wh1, wh2).prod(2) # [N,M]
return inter / (wh1.prod(2) + wh2.prod(2) - inter) # iou = inter / (area1 + area2 - inter)
if isinstance(boxes, torch.Tensor): # faster individually
boxes[:, 0].clamp_(0, shape[1]) # x1
boxes[:, 1].clamp_(0, shape[0]) # y1
boxes[:, 2].clamp_(0, shape[1]) # x2
boxes[:, 3].clamp_(0, shape[0]) # y2
else: # np.array (faster grouped)
boxes[:, [0, 2]] = boxes[:, [0, 2]].clip(0, shape[1]) # x1, x2
boxes[:, [1, 3]] = boxes[:, [1, 3]].clip(0, shape[0]) # y1, y2
def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, multi_label=False,
@@ -601,39 +747,48 @@ def strip_optimizer(f='best.pt', s=''): # from utils.general import *; strip_op
print(f"Optimizer stripped from {f},{(' saved as %s,' % s) if s else ''} {mb:.1f}MB")
def print_mutation(hyp, results, yaml_file='hyp_evolved.yaml', bucket=''):
# Print mutation results to evolve.txt (for use with train.py --evolve)
a = '%10s' * len(hyp) % tuple(hyp.keys()) # hyperparam keys
b = '%10.3g' * len(hyp) % tuple(hyp.values()) # hyperparam values
c = '%10.4g' * len(results) % results # results (P, R, mAP@0.5, mAP@0.5:0.95, val_losses x 3)
print('\n%s\n%s\nEvolved fitness: %s\n' % (a, b, c))
def print_mutation(results, hyp, save_dir, bucket):
evolve_csv, results_csv, evolve_yaml = save_dir / 'evolve.csv', save_dir / 'results.csv', save_dir / 'hyp_evolve.yaml'
keys = ('metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95',
'val/box_loss', 'val/obj_loss', 'val/cls_loss') + tuple(hyp.keys()) # [results + hyps]
keys = tuple(x.strip() for x in keys)
vals = results + tuple(hyp.values())
n = len(keys)
# Download (optional)
if bucket:
url = 'gs://%s/evolve.txt' % bucket
if gsutil_getsize(url) > (os.path.getsize('evolve.txt') if os.path.exists('evolve.txt') else 0):
os.system('gsutil cp %s .' % url) # download evolve.txt if larger than local
url = f'gs://{bucket}/evolve.csv'
if gsutil_getsize(url) > (os.path.getsize(evolve_csv) if os.path.exists(evolve_csv) else 0):
os.system(f'gsutil cp {url} {save_dir}') # download evolve.csv if larger than local
with open('evolve.txt', 'a') as f: # append result
f.write(c + b + '\n')
x = np.unique(np.loadtxt('evolve.txt', ndmin=2), axis=0) # load unique rows
x = x[np.argsort(-fitness(x))] # sort
np.savetxt('evolve.txt', x, '%10.3g') # save sort by fitness
# Log to evolve.csv
s = '' if evolve_csv.exists() else (('%20s,' * n % keys).rstrip(',') + '\n') # add header
with open(evolve_csv, 'a') as f:
f.write(s + ('%20.5g,' * n % vals).rstrip(',') + '\n')
# Print to screen
print(colorstr('evolve: ') + ', '.join(f'{x.strip():>20s}' for x in keys))
print(colorstr('evolve: ') + ', '.join(f'{x:20.5g}' for x in vals), end='\n\n\n')
# Save yaml
for i, k in enumerate(hyp.keys()):
hyp[k] = float(x[0, i + 7])
with open(yaml_file, 'w') as f:
results = tuple(x[0, :7])
c = '%10.4g' * len(results) % results # results (P, R, mAP@0.5, mAP@0.5:0.95, val_losses x 3)
f.write('# Hyperparameter Evolution Results\n# Generations: %g\n# Metrics: ' % len(x) + c + '\n\n')
with open(evolve_yaml, 'w') as f:
data = pd.read_csv(evolve_csv)
data = data.rename(columns=lambda x: x.strip()) # strip keys
i = np.argmax(fitness(data.values[:, :7])) #
f.write('# YOLOv3 Hyperparameter Evolution Results\n' +
f'# Best generation: {i}\n' +
f'# Last generation: {len(data)}\n' +
'# ' + ', '.join(f'{x.strip():>20s}' for x in keys[:7]) + '\n' +
'# ' + ', '.join(f'{x:>20.5g}' for x in data.values[i, :7]) + '\n\n')
yaml.safe_dump(hyp, f, sort_keys=False)
if bucket:
os.system('gsutil cp evolve.txt %s gs://%s' % (yaml_file, bucket)) # upload
os.system(f'gsutil cp {evolve_csv} {evolve_yaml} gs://{bucket}') # upload
def apply_classifier(x, model, img, im0):
# Apply a second stage classifier to yolo outputs
# Apply a second stage classifier to YOLO outputs
# Example model = torchvision.models.__dict__['efficientnet_b0'](pretrained=True).to(device).eval()
im0 = [im0] if isinstance(im0, np.ndarray) else im0
for i, d in enumerate(x): # per image
if d is not None and len(d):
@@ -654,11 +809,11 @@ def apply_classifier(x, model, img, im0):
for j, a in enumerate(d): # per item
cutout = im0[i][int(a[1]):int(a[3]), int(a[0]):int(a[2])]
im = cv2.resize(cutout, (224, 224)) # BGR
# cv2.imwrite('test%i.jpg' % j, cutout)
# cv2.imwrite('example%i.jpg' % j, cutout)
im = im[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
im = np.ascontiguousarray(im, dtype=np.float32) # uint8 to float32
im /= 255.0 # 0 - 255 to 0.0 - 1.0
im /= 255 # 0 - 255 to 0.0 - 1.0
ims.append(im)
pred_cls2 = model(torch.Tensor(ims).to(d.device)).argmax(1) # classifier prediction
@@ -667,33 +822,20 @@ def apply_classifier(x, model, img, im0):
return x
def save_one_box(xyxy, im, file='image.jpg', gain=1.02, pad=10, square=False, BGR=False, save=True):
# Save image crop as {file} with crop size multiple {gain} and {pad} pixels. Save and/or return crop
xyxy = torch.tensor(xyxy).view(-1, 4)
b = xyxy2xywh(xyxy) # boxes
if square:
b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1) # attempt rectangle to square
b[:, 2:] = b[:, 2:] * gain + pad # box wh * gain + pad
xyxy = xywh2xyxy(b).long()
clip_coords(xyxy, im.shape)
crop = im[int(xyxy[0, 1]):int(xyxy[0, 3]), int(xyxy[0, 0]):int(xyxy[0, 2]), ::(1 if BGR else -1)]
if save:
cv2.imwrite(str(increment_path(file, mkdir=True).with_suffix('.jpg')), crop)
return crop
def increment_path(path, exist_ok=False, sep='', mkdir=False):
# Increment file or directory path, i.e. runs/exp --> runs/exp{sep}2, runs/exp{sep}3, ... etc.
path = Path(path) # os-agnostic
if path.exists() and not exist_ok:
suffix = path.suffix
path = path.with_suffix('')
path, suffix = (path.with_suffix(''), path.suffix) if path.is_file() else (path, '')
dirs = glob.glob(f"{path}{sep}*") # similar paths
matches = [re.search(rf"%s{sep}(\d+)" % path.stem, d) for d in dirs]
i = [int(m.groups()[0]) for m in matches if m] # indices
n = max(i) + 1 if i else 2 # increment number
path = Path(f"{path}{sep}{n}{suffix}") # update path
dir = path if path.suffix == '' else path.parent # directory
if not dir.exists() and mkdir:
dir.mkdir(parents=True, exist_ok=True) # make directory
path = Path(f"{path}{sep}{n}{suffix}") # increment path
if mkdir:
path.mkdir(parents=True, exist_ok=True) # make directory
return path
# Variables
NCOLS = 0 if is_docker() else shutil.get_terminal_size().columns # terminal window size
-25
View File
@@ -1,25 +0,0 @@
FROM gcr.io/google-appengine/python
# Create a virtualenv for dependencies. This isolates these packages from
# system-level packages.
# Use -p python3 or -p python3.7 to select python version. Default is version 2.
RUN virtualenv /env -p python3
# Setting these environment variables are the same as running
# source /env/bin/activate.
ENV VIRTUAL_ENV /env
ENV PATH /env/bin:$PATH
RUN apt-get update && apt-get install -y python-opencv
# Copy the application's requirements.txt and run pip to install all
# dependencies into the virtualenv.
ADD requirements.txt /app/requirements.txt
RUN pip install -r /app/requirements.txt
# Add the application source code.
ADD . /app
# Run a WSGI server to serve the application. gunicorn must be declared as
# a dependency in requirements.txt.
CMD gunicorn -b :$PORT main:app
@@ -1,4 +0,0 @@
# add these requirements in your app on top of the existing ones
pip==19.2
Flask==1.0.2
gunicorn==19.9.0
-14
View File
@@ -1,14 +0,0 @@
runtime: custom
env: flex
service: yolov3app
liveness_check:
initial_delay_sec: 600
manual_scaling:
instances: 1
resources:
cpu: 1
memory_gb: 4
disk_size_gb: 20
+156
View File
@@ -0,0 +1,156 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Logging utils
"""
import os
import warnings
from threading import Thread
import pkg_resources as pkg
import torch
from torch.utils.tensorboard import SummaryWriter
from utils.general import colorstr, emojis
from utils.loggers.wandb.wandb_utils import WandbLogger
from utils.plots import plot_images, plot_results
from utils.torch_utils import de_parallel
LOGGERS = ('csv', 'tb', 'wandb') # text-file, TensorBoard, Weights & Biases
RANK = int(os.getenv('RANK', -1))
try:
import wandb
assert hasattr(wandb, '__version__') # verify package import not local dir
if pkg.parse_version(wandb.__version__) >= pkg.parse_version('0.12.2') and RANK in [0, -1]:
wandb_login_success = wandb.login(timeout=30)
if not wandb_login_success:
wandb = None
except (ImportError, AssertionError):
wandb = None
class Loggers():
# Loggers class
def __init__(self, save_dir=None, weights=None, opt=None, hyp=None, logger=None, include=LOGGERS):
self.save_dir = save_dir
self.weights = weights
self.opt = opt
self.hyp = hyp
self.logger = logger # for printing results to console
self.include = include
self.keys = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', # train loss
'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', # metrics
'val/box_loss', 'val/obj_loss', 'val/cls_loss', # val loss
'x/lr0', 'x/lr1', 'x/lr2'] # params
for k in LOGGERS:
setattr(self, k, None) # init empty logger dictionary
self.csv = True # always log to csv
# Message
if not wandb:
prefix = colorstr('Weights & Biases: ')
s = f"{prefix}run 'pip install wandb' to automatically track and visualize YOLOv3 🚀 runs (RECOMMENDED)"
print(emojis(s))
# TensorBoard
s = self.save_dir
if 'tb' in self.include and not self.opt.evolve:
prefix = colorstr('TensorBoard: ')
self.logger.info(f"{prefix}Start with 'tensorboard --logdir {s.parent}', view at http://localhost:6006/")
self.tb = SummaryWriter(str(s))
# W&B
if wandb and 'wandb' in self.include:
wandb_artifact_resume = isinstance(self.opt.resume, str) and self.opt.resume.startswith('wandb-artifact://')
run_id = torch.load(self.weights).get('wandb_id') if self.opt.resume and not wandb_artifact_resume else None
self.opt.hyp = self.hyp # add hyperparameters
self.wandb = WandbLogger(self.opt, run_id)
else:
self.wandb = None
def on_pretrain_routine_end(self):
# Callback runs on pre-train routine end
paths = self.save_dir.glob('*labels*.jpg') # training labels
if self.wandb:
self.wandb.log({"Labels": [wandb.Image(str(x), caption=x.name) for x in paths]})
def on_train_batch_end(self, ni, model, imgs, targets, paths, plots, sync_bn):
# Callback runs on train batch end
if plots:
if ni == 0:
if not sync_bn: # tb.add_graph() --sync known issue https://github.com/ultralytics/yolov5/issues/3754
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress jit trace warning
self.tb.add_graph(torch.jit.trace(de_parallel(model), imgs[0:1], strict=False), [])
if ni < 3:
f = self.save_dir / f'train_batch{ni}.jpg' # filename
Thread(target=plot_images, args=(imgs, targets, paths, f), daemon=True).start()
if self.wandb and ni == 10:
files = sorted(self.save_dir.glob('train*.jpg'))
self.wandb.log({'Mosaics': [wandb.Image(str(f), caption=f.name) for f in files if f.exists()]})
def on_train_epoch_end(self, epoch):
# Callback runs on train epoch end
if self.wandb:
self.wandb.current_epoch = epoch + 1
def on_val_image_end(self, pred, predn, path, names, im):
# Callback runs on val image end
if self.wandb:
self.wandb.val_one_image(pred, predn, path, names, im)
def on_val_end(self):
# Callback runs on val end
if self.wandb:
files = sorted(self.save_dir.glob('val*.jpg'))
self.wandb.log({"Validation": [wandb.Image(str(f), caption=f.name) for f in files]})
def on_fit_epoch_end(self, vals, epoch, best_fitness, fi):
# Callback runs at the end of each fit (train+val) epoch
x = {k: v for k, v in zip(self.keys, vals)} # dict
if self.csv:
file = self.save_dir / 'results.csv'
n = len(x) + 1 # number of cols
s = '' if file.exists() else (('%20s,' * n % tuple(['epoch'] + self.keys)).rstrip(',') + '\n') # add header
with open(file, 'a') as f:
f.write(s + ('%20.5g,' * n % tuple([epoch] + vals)).rstrip(',') + '\n')
if self.tb:
for k, v in x.items():
self.tb.add_scalar(k, v, epoch)
if self.wandb:
self.wandb.log(x)
self.wandb.end_epoch(best_result=best_fitness == fi)
def on_model_save(self, last, epoch, final_epoch, best_fitness, fi):
# Callback runs on model save event
if self.wandb:
if ((epoch + 1) % self.opt.save_period == 0 and not final_epoch) and self.opt.save_period != -1:
self.wandb.log_model(last.parent, self.opt, epoch, fi, best_model=best_fitness == fi)
def on_train_end(self, last, best, plots, epoch, results):
# Callback runs on training end
if plots:
plot_results(file=self.save_dir / 'results.csv') # save results.png
files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))]
files = [(self.save_dir / f) for f in files if (self.save_dir / f).exists()] # filter
if self.tb:
import cv2
for f in files:
self.tb.add_image(f.stem, cv2.imread(str(f))[..., ::-1], epoch, dataformats='HWC')
if self.wandb:
self.wandb.log({"Results": [wandb.Image(str(f), caption=f.name) for f in files]})
# Calling wandb.log. TODO: Refactor this into WandbLogger.log_model
if not self.opt.evolve:
wandb.log_artifact(str(best if best.exists() else last), type='model',
name='run_' + self.wandb.wandb_run.id + '_model',
aliases=['latest', 'best', 'stripped'])
self.wandb.finish_run()
else:
self.wandb.finish_run()
self.wandb = WandbLogger(self.opt)
+147
View File
@@ -0,0 +1,147 @@
📚 This guide explains how to use **Weights & Biases** (W&B) with YOLOv3 🚀. UPDATED 29 September 2021.
* [About Weights & Biases](#about-weights-&-biases)
* [First-Time Setup](#first-time-setup)
* [Viewing runs](#viewing-runs)
* [Advanced Usage: Dataset Versioning and Evaluation](#advanced-usage)
* [Reports: Share your work with the world!](#reports)
## About Weights & Biases
Think of [W&B](https://wandb.ai/site?utm_campaign=repo_yolo_wandbtutorial) like GitHub for machine learning models. With a few lines of code, save everything you need to debug, compare and reproduce your models — architecture, hyperparameters, git commits, model weights, GPU usage, and even datasets and predictions.
Used by top researchers including teams at OpenAI, Lyft, Github, and MILA, W&B is part of the new standard of best practices for machine learning. How W&B can help you optimize your machine learning workflows:
* [Debug](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#Free-2) model performance in real time
* [GPU usage](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#System-4) visualized automatically
* [Custom charts](https://wandb.ai/wandb/customizable-charts/reports/Powerful-Custom-Charts-To-Debug-Model-Peformance--VmlldzoyNzY4ODI) for powerful, extensible visualization
* [Share insights](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#Share-8) interactively with collaborators
* [Optimize hyperparameters](https://docs.wandb.com/sweeps) efficiently
* [Track](https://docs.wandb.com/artifacts) datasets, pipelines, and production models
## First-Time Setup
<details open>
<summary> Toggle Details </summary>
When you first train, W&B will prompt you to create a new account and will generate an **API key** for you. If you are an existing user you can retrieve your key from https://wandb.ai/authorize. This key is used to tell W&B where to log your data. You only need to supply your key once, and then it is remembered on the same device.
W&B will create a cloud **project** (default is 'YOLOv3') for your training runs, and each new training run will be provided a unique run **name** within that project as project/name. You can also manually set your project and run name as:
```shell
$ python train.py --project ... --name ...
```
YOLOv3 notebook example: <a href="https://colab.research.google.com/github/ultralytics/yolov3/blob/master/tutorial.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"></a> <a href="https://www.kaggle.com/ultralytics/yolov3"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a>
<img width="960" alt="Screen Shot 2021-09-29 at 10 23 13 PM" src="https://user-images.githubusercontent.com/26833433/135392431-1ab7920a-c49d-450a-b0b0-0c86ec86100e.png">
</details>
## Viewing Runs
<details open>
<summary> Toggle Details </summary>
Run information streams from your environment to the W&B cloud console as you train. This allows you to monitor and even cancel runs in <b>realtime</b> . All important information is logged:
* Training & Validation losses
* Metrics: Precision, Recall, mAP@0.5, mAP@0.5:0.95
* Learning Rate over time
* A bounding box debugging panel, showing the training progress over time
* GPU: Type, **GPU Utilization**, power, temperature, **CUDA memory usage**
* System: Disk I/0, CPU utilization, RAM memory usage
* Your trained model as W&B Artifact
* Environment: OS and Python types, Git repository and state, **training command**
<p align="center"><img width="900" alt="Weights & Biases dashboard" src="https://user-images.githubusercontent.com/26833433/135390767-c28b050f-8455-4004-adb0-3b730386e2b2.png"></p>
</details>
## Advanced Usage
You can leverage W&B artifacts and Tables integration to easily visualize and manage your datasets, models and training evaluations. Here are some quick examples to get you started.
<details open>
<h3>1. Visualize and Version Datasets</h3>
Log, visualize, dynamically query, and understand your data with <a href='https://docs.wandb.ai/guides/data-vis/tables'>W&B Tables</a>. You can use the following command to log your dataset as a W&B Table. This will generate a <code>{dataset}_wandb.yaml</code> file which can be used to train from dataset artifact.
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python utils/logger/wandb/log_dataset.py --project ... --name ... --data .. </code>
![Screenshot (64)](https://user-images.githubusercontent.com/15766192/128486078-d8433890-98a3-4d12-8986-b6c0e3fc64b9.png)
</details>
<h3> 2: Train and Log Evaluation simultaneousy </h3>
This is an extension of the previous section, but it'll also training after uploading the dataset. <b> This also evaluation Table</b>
Evaluation table compares your predictions and ground truths across the validation set for each epoch. It uses the references to the already uploaded datasets,
so no images will be uploaded from your system more than once.
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python utils/logger/wandb/log_dataset.py --data .. --upload_data </code>
![Screenshot (72)](https://user-images.githubusercontent.com/15766192/128979739-4cf63aeb-a76f-483f-8861-1c0100b938a5.png)
</details>
<h3> 3: Train using dataset artifact </h3>
When you upload a dataset as described in the first section, you get a new config file with an added `_wandb` to its name. This file contains the information that
can be used to train a model directly from the dataset artifact. <b> This also logs evaluation </b>
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python utils/logger/wandb/log_dataset.py --data {data}_wandb.yaml </code>
![Screenshot (72)](https://user-images.githubusercontent.com/15766192/128979739-4cf63aeb-a76f-483f-8861-1c0100b938a5.png)
</details>
<h3> 4: Save model checkpoints as artifacts </h3>
To enable saving and versioning checkpoints of your experiment, pass `--save_period n` with the base cammand, where `n` represents checkpoint interval.
You can also log both the dataset and model checkpoints simultaneously. If not passed, only the final model will be logged
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --save_period 1 </code>
![Screenshot (68)](https://user-images.githubusercontent.com/15766192/128726138-ec6c1f60-639d-437d-b4ee-3acd9de47ef3.png)
</details>
</details>
<h3> 5: Resume runs from checkpoint artifacts. </h3>
Any run can be resumed using artifacts if the <code>--resume</code> argument starts with <code>wandb-artifact://</code> prefix followed by the run path, i.e, <code>wandb-artifact://username/project/runid </code>. This doesn't require the model checkpoint to be present on the local system.
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --resume wandb-artifact://{run_path} </code>
![Screenshot (70)](https://user-images.githubusercontent.com/15766192/128728988-4e84b355-6c87-41ae-a591-14aecf45343e.png)
</details>
<h3> 6: Resume runs from dataset artifact & checkpoint artifacts. </h3>
<b> Local dataset or model checkpoints are not required. This can be used to resume runs directly on a different device </b>
The syntax is same as the previous section, but you'll need to lof both the dataset and model checkpoints as artifacts, i.e, set bot <code>--upload_dataset</code> or
train from <code>_wandb.yaml</code> file and set <code>--save_period</code>
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --resume wandb-artifact://{run_path} </code>
![Screenshot (70)](https://user-images.githubusercontent.com/15766192/128728988-4e84b355-6c87-41ae-a591-14aecf45343e.png)
</details>
</details>
<h3> Reports </h3>
W&B Reports can be created from your saved runs for sharing online. Once a report is created you will receive a link you can use to publically share your results. Here is an example report created from the COCO128 tutorial trainings of all YOLOv5 models ([link](https://wandb.ai/glenn-jocher/yolov5_tutorial/reports/YOLOv5-COCO128-Tutorial-Results--VmlldzozMDI5OTY)).
<img width="900" alt="Weights & Biases Reports" src="https://user-images.githubusercontent.com/26833433/135394029-a17eaf86-c6c1-4b1d-bb80-b90e83aaffa7.png">
## Environments
YOLOv3 may be run in any of the following up-to-date verified environments (with all dependencies including [CUDA](https://developer.nvidia.com/cuda)/[CUDNN](https://developer.nvidia.com/cudnn), [Python](https://www.python.org/) and [PyTorch](https://pytorch.org/) preinstalled):
- **Google Colab and Kaggle** notebooks with free GPU: <a href="https://colab.research.google.com/github/ultralytics/yolov3/blob/master/tutorial.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"></a> <a href="https://www.kaggle.com/ultralytics/yolov3"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a>
- **Google Cloud** Deep Learning VM. See [GCP Quickstart Guide](https://github.com/ultralytics/yolov3/wiki/GCP-Quickstart)
- **Amazon** Deep Learning AMI. See [AWS Quickstart Guide](https://github.com/ultralytics/yolov3/wiki/AWS-Quickstart)
- **Docker Image**. See [Docker Quickstart Guide](https://github.com/ultralytics/yolov3/wiki/Docker-Quickstart) <a href="https://hub.docker.com/r/ultralytics/yolov3"><img src="https://img.shields.io/docker/pulls/ultralytics/yolov3?logo=docker" alt="Docker Pulls"></a>
## Status
![CI CPU testing](https://github.com/ultralytics/yolov3/workflows/CI%20CPU%20testing/badge.svg)
If this badge is green, all [YOLOv3 GitHub Actions](https://github.com/ultralytics/yolov3/actions) Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv3 training ([train.py](https://github.com/ultralytics/yolov3/blob/master/train.py)), validation ([val.py](https://github.com/ultralytics/yolov3/blob/master/val.py)), inference ([detect.py](https://github.com/ultralytics/yolov3/blob/master/detect.py)) and export ([export.py](https://github.com/ultralytics/yolov3/blob/master/export.py)) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.
@@ -1,16 +1,16 @@
import argparse
import yaml
from wandb_utils import WandbLogger
from utils.general import LOGGER
WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
def create_dataset_artifact(opt):
with open(opt.data) as f:
data = yaml.safe_load(f) # data dict
logger = WandbLogger(opt, '', None, data, job_type='Dataset Creation')
logger = WandbLogger(opt, None, job_type='Dataset Creation') # TODO: return value unused
if not logger.wandb:
LOGGER.info("install wandb using `pip install wandb` to log the dataset")
if __name__ == '__main__':
@@ -18,6 +18,9 @@ if __name__ == '__main__':
parser.add_argument('--data', type=str, default='data/coco128.yaml', help='data.yaml path')
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
parser.add_argument('--project', type=str, default='YOLOv3', help='name of W&B Project')
parser.add_argument('--entity', default=None, help='W&B entity')
parser.add_argument('--name', type=str, default='log dataset', help='name of W&B run')
opt = parser.parse_args()
opt.resume = False # Explicitly disallow resume check for dataset upload job
+41
View File
@@ -0,0 +1,41 @@
import sys
from pathlib import Path
import wandb
FILE = Path(__file__).resolve()
ROOT = FILE.parents[3] # root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
from train import parse_opt, train
from utils.callbacks import Callbacks
from utils.general import increment_path
from utils.torch_utils import select_device
def sweep():
wandb.init()
# Get hyp dict from sweep agent
hyp_dict = vars(wandb.config).get("_items")
# Workaround: get necessary opt args
opt = parse_opt(known=True)
opt.batch_size = hyp_dict.get("batch_size")
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok or opt.evolve))
opt.epochs = hyp_dict.get("epochs")
opt.nosave = True
opt.data = hyp_dict.get("data")
opt.weights = str(opt.weights)
opt.cfg = str(opt.cfg)
opt.data = str(opt.data)
opt.hyp = str(opt.hyp)
opt.project = str(opt.project)
device = select_device(opt.device, batch_size=opt.batch_size)
# train
train(hyp_dict, opt, device, callbacks=Callbacks())
if __name__ == "__main__":
sweep()
+143
View File
@@ -0,0 +1,143 @@
# Hyperparameters for training
# To set range-
# Provide min and max values as:
# parameter:
#
# min: scalar
# max: scalar
# OR
#
# Set a specific list of search space-
# parameter:
# values: [scalar1, scalar2, scalar3...]
#
# You can use grid, bayesian and hyperopt search strategy
# For more info on configuring sweeps visit - https://docs.wandb.ai/guides/sweeps/configuration
program: utils/loggers/wandb/sweep.py
method: random
metric:
name: metrics/mAP_0.5
goal: maximize
parameters:
# hyperparameters: set either min, max range or values list
data:
value: "data/coco128.yaml"
batch_size:
values: [64]
epochs:
values: [10]
lr0:
distribution: uniform
min: 1e-5
max: 1e-1
lrf:
distribution: uniform
min: 0.01
max: 1.0
momentum:
distribution: uniform
min: 0.6
max: 0.98
weight_decay:
distribution: uniform
min: 0.0
max: 0.001
warmup_epochs:
distribution: uniform
min: 0.0
max: 5.0
warmup_momentum:
distribution: uniform
min: 0.0
max: 0.95
warmup_bias_lr:
distribution: uniform
min: 0.0
max: 0.2
box:
distribution: uniform
min: 0.02
max: 0.2
cls:
distribution: uniform
min: 0.2
max: 4.0
cls_pw:
distribution: uniform
min: 0.5
max: 2.0
obj:
distribution: uniform
min: 0.2
max: 4.0
obj_pw:
distribution: uniform
min: 0.5
max: 2.0
iou_t:
distribution: uniform
min: 0.1
max: 0.7
anchor_t:
distribution: uniform
min: 2.0
max: 8.0
fl_gamma:
distribution: uniform
min: 0.0
max: 0.1
hsv_h:
distribution: uniform
min: 0.0
max: 0.1
hsv_s:
distribution: uniform
min: 0.0
max: 0.9
hsv_v:
distribution: uniform
min: 0.0
max: 0.9
degrees:
distribution: uniform
min: 0.0
max: 45.0
translate:
distribution: uniform
min: 0.0
max: 0.9
scale:
distribution: uniform
min: 0.0
max: 0.9
shear:
distribution: uniform
min: 0.0
max: 10.0
perspective:
distribution: uniform
min: 0.0
max: 0.001
flipud:
distribution: uniform
min: 0.0
max: 1.0
fliplr:
distribution: uniform
min: 0.0
max: 1.0
mosaic:
distribution: uniform
min: 0.0
max: 1.0
mixup:
distribution: uniform
min: 0.0
max: 1.0
copy_paste:
distribution: uniform
min: 0.0
max: 1.0
+532
View File
@@ -0,0 +1,532 @@
"""Utilities and tools for tracking runs with Weights & Biases."""
import logging
import os
import sys
from contextlib import contextmanager
from pathlib import Path
from typing import Dict
import pkg_resources as pkg
import yaml
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[3] # root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
from utils.datasets import LoadImagesAndLabels, img2label_paths
from utils.general import LOGGER, check_dataset, check_file
try:
import wandb
assert hasattr(wandb, '__version__') # verify package import not local dir
except (ImportError, AssertionError):
wandb = None
RANK = int(os.getenv('RANK', -1))
WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
def remove_prefix(from_string, prefix=WANDB_ARTIFACT_PREFIX):
return from_string[len(prefix):]
def check_wandb_config_file(data_config_file):
wandb_config = '_wandb.'.join(data_config_file.rsplit('.', 1)) # updated data.yaml path
if Path(wandb_config).is_file():
return wandb_config
return data_config_file
def check_wandb_dataset(data_file):
is_trainset_wandb_artifact = False
is_valset_wandb_artifact = False
if check_file(data_file) and data_file.endswith('.yaml'):
with open(data_file, errors='ignore') as f:
data_dict = yaml.safe_load(f)
is_trainset_wandb_artifact = (isinstance(data_dict['train'], str) and
data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX))
is_valset_wandb_artifact = (isinstance(data_dict['val'], str) and
data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX))
if is_trainset_wandb_artifact or is_valset_wandb_artifact:
return data_dict
else:
return check_dataset(data_file)
def get_run_info(run_path):
run_path = Path(remove_prefix(run_path, WANDB_ARTIFACT_PREFIX))
run_id = run_path.stem
project = run_path.parent.stem
entity = run_path.parent.parent.stem
model_artifact_name = 'run_' + run_id + '_model'
return entity, project, run_id, model_artifact_name
def check_wandb_resume(opt):
process_wandb_config_ddp_mode(opt) if RANK not in [-1, 0] else None
if isinstance(opt.resume, str):
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
if RANK not in [-1, 0]: # For resuming DDP runs
entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
api = wandb.Api()
artifact = api.artifact(entity + '/' + project + '/' + model_artifact_name + ':latest')
modeldir = artifact.download()
opt.weights = str(Path(modeldir) / "last.pt")
return True
return None
def process_wandb_config_ddp_mode(opt):
with open(check_file(opt.data), errors='ignore') as f:
data_dict = yaml.safe_load(f) # data dict
train_dir, val_dir = None, None
if isinstance(data_dict['train'], str) and data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX):
api = wandb.Api()
train_artifact = api.artifact(remove_prefix(data_dict['train']) + ':' + opt.artifact_alias)
train_dir = train_artifact.download()
train_path = Path(train_dir) / 'data/images/'
data_dict['train'] = str(train_path)
if isinstance(data_dict['val'], str) and data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX):
api = wandb.Api()
val_artifact = api.artifact(remove_prefix(data_dict['val']) + ':' + opt.artifact_alias)
val_dir = val_artifact.download()
val_path = Path(val_dir) / 'data/images/'
data_dict['val'] = str(val_path)
if train_dir or val_dir:
ddp_data_path = str(Path(val_dir) / 'wandb_local_data.yaml')
with open(ddp_data_path, 'w') as f:
yaml.safe_dump(data_dict, f)
opt.data = ddp_data_path
class WandbLogger():
"""Log training runs, datasets, models, and predictions to Weights & Biases.
This logger sends information to W&B at wandb.ai. By default, this information
includes hyperparameters, system configuration and metrics, model metrics,
and basic data metrics and analyses.
By providing additional command line arguments to train.py, datasets,
models and predictions can also be logged.
For more on how this logger is used, see the Weights & Biases documentation:
https://docs.wandb.com/guides/integrations/yolov5
"""
def __init__(self, opt, run_id=None, job_type='Training'):
"""
- Initialize WandbLogger instance
- Upload dataset if opt.upload_dataset is True
- Setup trainig processes if job_type is 'Training'
arguments:
opt (namespace) -- Commandline arguments for this run
run_id (str) -- Run ID of W&B run to be resumed
job_type (str) -- To set the job_type for this run
"""
# Pre-training routine --
self.job_type = job_type
self.wandb, self.wandb_run = wandb, None if not wandb else wandb.run
self.val_artifact, self.train_artifact = None, None
self.train_artifact_path, self.val_artifact_path = None, None
self.result_artifact = None
self.val_table, self.result_table = None, None
self.bbox_media_panel_images = []
self.val_table_path_map = None
self.max_imgs_to_log = 16
self.wandb_artifact_data_dict = None
self.data_dict = None
# It's more elegant to stick to 1 wandb.init call,
# but useful config data is overwritten in the WandbLogger's wandb.init call
if isinstance(opt.resume, str): # checks resume from artifact
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
model_artifact_name = WANDB_ARTIFACT_PREFIX + model_artifact_name
assert wandb, 'install wandb to resume wandb runs'
# Resume wandb-artifact:// runs here| workaround for not overwriting wandb.config
self.wandb_run = wandb.init(id=run_id,
project=project,
entity=entity,
resume='allow',
allow_val_change=True)
opt.resume = model_artifact_name
elif self.wandb:
self.wandb_run = wandb.init(config=opt,
resume="allow",
project='YOLOv3' if opt.project == 'runs/train' else Path(opt.project).stem,
entity=opt.entity,
name=opt.name if opt.name != 'exp' else None,
job_type=job_type,
id=run_id,
allow_val_change=True) if not wandb.run else wandb.run
if self.wandb_run:
if self.job_type == 'Training':
if opt.upload_dataset:
if not opt.resume:
self.wandb_artifact_data_dict = self.check_and_upload_dataset(opt)
if opt.resume:
# resume from artifact
if isinstance(opt.resume, str) and opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
self.data_dict = dict(self.wandb_run.config.data_dict)
else: # local resume
self.data_dict = check_wandb_dataset(opt.data)
else:
self.data_dict = check_wandb_dataset(opt.data)
self.wandb_artifact_data_dict = self.wandb_artifact_data_dict or self.data_dict
# write data_dict to config. useful for resuming from artifacts. Do this only when not resuming.
self.wandb_run.config.update({'data_dict': self.wandb_artifact_data_dict},
allow_val_change=True)
self.setup_training(opt)
if self.job_type == 'Dataset Creation':
self.data_dict = self.check_and_upload_dataset(opt)
def check_and_upload_dataset(self, opt):
"""
Check if the dataset format is compatible and upload it as W&B artifact
arguments:
opt (namespace)-- Commandline arguments for current run
returns:
Updated dataset info dictionary where local dataset paths are replaced by WAND_ARFACT_PREFIX links.
"""
assert wandb, 'Install wandb to upload dataset'
config_path = self.log_dataset_artifact(opt.data,
opt.single_cls,
'YOLOv3' if opt.project == 'runs/train' else Path(opt.project).stem)
LOGGER.info(f"Created dataset config file {config_path}")
with open(config_path, errors='ignore') as f:
wandb_data_dict = yaml.safe_load(f)
return wandb_data_dict
def setup_training(self, opt):
"""
Setup the necessary processes for training YOLO models:
- Attempt to download model checkpoint and dataset artifacts if opt.resume stats with WANDB_ARTIFACT_PREFIX
- Update data_dict, to contain info of previous run if resumed and the paths of dataset artifact if downloaded
- Setup log_dict, initialize bbox_interval
arguments:
opt (namespace) -- commandline arguments for this run
"""
self.log_dict, self.current_epoch = {}, 0
self.bbox_interval = opt.bbox_interval
if isinstance(opt.resume, str):
modeldir, _ = self.download_model_artifact(opt)
if modeldir:
self.weights = Path(modeldir) / "last.pt"
config = self.wandb_run.config
opt.weights, opt.save_period, opt.batch_size, opt.bbox_interval, opt.epochs, opt.hyp = str(
self.weights), config.save_period, config.batch_size, config.bbox_interval, config.epochs, \
config.hyp
data_dict = self.data_dict
if self.val_artifact is None: # If --upload_dataset is set, use the existing artifact, don't download
self.train_artifact_path, self.train_artifact = self.download_dataset_artifact(data_dict.get('train'),
opt.artifact_alias)
self.val_artifact_path, self.val_artifact = self.download_dataset_artifact(data_dict.get('val'),
opt.artifact_alias)
if self.train_artifact_path is not None:
train_path = Path(self.train_artifact_path) / 'data/images/'
data_dict['train'] = str(train_path)
if self.val_artifact_path is not None:
val_path = Path(self.val_artifact_path) / 'data/images/'
data_dict['val'] = str(val_path)
if self.val_artifact is not None:
self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
self.result_table = wandb.Table(["epoch", "id", "ground truth", "prediction", "avg_confidence"])
self.val_table = self.val_artifact.get("val")
if self.val_table_path_map is None:
self.map_val_table_path()
if opt.bbox_interval == -1:
self.bbox_interval = opt.bbox_interval = (opt.epochs // 10) if opt.epochs > 10 else 1
train_from_artifact = self.train_artifact_path is not None and self.val_artifact_path is not None
# Update the the data_dict to point to local artifacts dir
if train_from_artifact:
self.data_dict = data_dict
def download_dataset_artifact(self, path, alias):
"""
download the model checkpoint artifact if the path starts with WANDB_ARTIFACT_PREFIX
arguments:
path -- path of the dataset to be used for training
alias (str)-- alias of the artifact to be download/used for training
returns:
(str, wandb.Artifact) -- path of the downladed dataset and it's corresponding artifact object if dataset
is found otherwise returns (None, None)
"""
if isinstance(path, str) and path.startswith(WANDB_ARTIFACT_PREFIX):
artifact_path = Path(remove_prefix(path, WANDB_ARTIFACT_PREFIX) + ":" + alias)
dataset_artifact = wandb.use_artifact(artifact_path.as_posix().replace("\\", "/"))
assert dataset_artifact is not None, "'Error: W&B dataset artifact doesn\'t exist'"
datadir = dataset_artifact.download()
return datadir, dataset_artifact
return None, None
def download_model_artifact(self, opt):
"""
download the model checkpoint artifact if the resume path starts with WANDB_ARTIFACT_PREFIX
arguments:
opt (namespace) -- Commandline arguments for this run
"""
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
model_artifact = wandb.use_artifact(remove_prefix(opt.resume, WANDB_ARTIFACT_PREFIX) + ":latest")
assert model_artifact is not None, 'Error: W&B model artifact doesn\'t exist'
modeldir = model_artifact.download()
epochs_trained = model_artifact.metadata.get('epochs_trained')
total_epochs = model_artifact.metadata.get('total_epochs')
is_finished = total_epochs is None
assert not is_finished, 'training is finished, can only resume incomplete runs.'
return modeldir, model_artifact
return None, None
def log_model(self, path, opt, epoch, fitness_score, best_model=False):
"""
Log the model checkpoint as W&B artifact
arguments:
path (Path) -- Path of directory containing the checkpoints
opt (namespace) -- Command line arguments for this run
epoch (int) -- Current epoch number
fitness_score (float) -- fitness score for current epoch
best_model (boolean) -- Boolean representing if the current checkpoint is the best yet.
"""
model_artifact = wandb.Artifact('run_' + wandb.run.id + '_model', type='model', metadata={
'original_url': str(path),
'epochs_trained': epoch + 1,
'save period': opt.save_period,
'project': opt.project,
'total_epochs': opt.epochs,
'fitness_score': fitness_score
})
model_artifact.add_file(str(path / 'last.pt'), name='last.pt')
wandb.log_artifact(model_artifact,
aliases=['latest', 'last', 'epoch ' + str(self.current_epoch), 'best' if best_model else ''])
LOGGER.info(f"Saving model artifact on epoch {epoch + 1}")
def log_dataset_artifact(self, data_file, single_cls, project, overwrite_config=False):
"""
Log the dataset as W&B artifact and return the new data file with W&B links
arguments:
data_file (str) -- the .yaml file with information about the dataset like - path, classes etc.
single_class (boolean) -- train multi-class data as single-class
project (str) -- project name. Used to construct the artifact path
overwrite_config (boolean) -- overwrites the data.yaml file if set to true otherwise creates a new
file with _wandb postfix. Eg -> data_wandb.yaml
returns:
the new .yaml file with artifact links. it can be used to start training directly from artifacts
"""
self.data_dict = check_dataset(data_file) # parse and check
data = dict(self.data_dict)
nc, names = (1, ['item']) if single_cls else (int(data['nc']), data['names'])
names = {k: v for k, v in enumerate(names)} # to index dictionary
self.train_artifact = self.create_dataset_table(LoadImagesAndLabels(
data['train'], rect=True, batch_size=1), names, name='train') if data.get('train') else None
self.val_artifact = self.create_dataset_table(LoadImagesAndLabels(
data['val'], rect=True, batch_size=1), names, name='val') if data.get('val') else None
if data.get('train'):
data['train'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'train')
if data.get('val'):
data['val'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'val')
path = Path(data_file).stem
path = (path if overwrite_config else path + '_wandb') + '.yaml' # updated data.yaml path
data.pop('download', None)
data.pop('path', None)
with open(path, 'w') as f:
yaml.safe_dump(data, f)
if self.job_type == 'Training': # builds correct artifact pipeline graph
self.wandb_run.use_artifact(self.val_artifact)
self.wandb_run.use_artifact(self.train_artifact)
self.val_artifact.wait()
self.val_table = self.val_artifact.get('val')
self.map_val_table_path()
else:
self.wandb_run.log_artifact(self.train_artifact)
self.wandb_run.log_artifact(self.val_artifact)
return path
def map_val_table_path(self):
"""
Map the validation dataset Table like name of file -> it's id in the W&B Table.
Useful for - referencing artifacts for evaluation.
"""
self.val_table_path_map = {}
LOGGER.info("Mapping dataset")
for i, data in enumerate(tqdm(self.val_table.data)):
self.val_table_path_map[data[3]] = data[0]
def create_dataset_table(self, dataset: LoadImagesAndLabels, class_to_id: Dict[int,str], name: str = 'dataset'):
"""
Create and return W&B artifact containing W&B Table of the dataset.
arguments:
dataset -- instance of LoadImagesAndLabels class used to iterate over the data to build Table
class_to_id -- hash map that maps class ids to labels
name -- name of the artifact
returns:
dataset artifact to be logged or used
"""
# TODO: Explore multiprocessing to slpit this loop parallely| This is essential for speeding up the the logging
artifact = wandb.Artifact(name=name, type="dataset")
img_files = tqdm([dataset.path]) if isinstance(dataset.path, str) and Path(dataset.path).is_dir() else None
img_files = tqdm(dataset.img_files) if not img_files else img_files
for img_file in img_files:
if Path(img_file).is_dir():
artifact.add_dir(img_file, name='data/images')
labels_path = 'labels'.join(dataset.path.rsplit('images', 1))
artifact.add_dir(labels_path, name='data/labels')
else:
artifact.add_file(img_file, name='data/images/' + Path(img_file).name)
label_file = Path(img2label_paths([img_file])[0])
artifact.add_file(str(label_file),
name='data/labels/' + label_file.name) if label_file.exists() else None
table = wandb.Table(columns=["id", "train_image", "Classes", "name"])
class_set = wandb.Classes([{'id': id, 'name': name} for id, name in class_to_id.items()])
for si, (img, labels, paths, shapes) in enumerate(tqdm(dataset)):
box_data, img_classes = [], {}
for cls, *xywh in labels[:, 1:].tolist():
cls = int(cls)
box_data.append({"position": {"middle": [xywh[0], xywh[1]], "width": xywh[2], "height": xywh[3]},
"class_id": cls,
"box_caption": "%s" % (class_to_id[cls])})
img_classes[cls] = class_to_id[cls]
boxes = {"ground_truth": {"box_data": box_data, "class_labels": class_to_id}} # inference-space
table.add_data(si, wandb.Image(paths, classes=class_set, boxes=boxes), list(img_classes.values()),
Path(paths).name)
artifact.add(table, name)
return artifact
def log_training_progress(self, predn, path, names):
"""
Build evaluation Table. Uses reference from validation dataset table.
arguments:
predn (list): list of predictions in the native space in the format - [xmin, ymin, xmax, ymax, confidence, class]
path (str): local path of the current evaluation image
names (dict(int, str)): hash map that maps class ids to labels
"""
class_set = wandb.Classes([{'id': id, 'name': name} for id, name in names.items()])
box_data = []
total_conf = 0
for *xyxy, conf, cls in predn.tolist():
if conf >= 0.25:
box_data.append(
{"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]},
"class_id": int(cls),
"box_caption": f"{names[cls]} {conf:.3f}",
"scores": {"class_score": conf},
"domain": "pixel"})
total_conf += conf
boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
id = self.val_table_path_map[Path(path).name]
self.result_table.add_data(self.current_epoch,
id,
self.val_table.data[id][1],
wandb.Image(self.val_table.data[id][1], boxes=boxes, classes=class_set),
total_conf / max(1, len(box_data))
)
def val_one_image(self, pred, predn, path, names, im):
"""
Log validation data for one image. updates the result Table if validation dataset is uploaded and log bbox media panel
arguments:
pred (list): list of scaled predictions in the format - [xmin, ymin, xmax, ymax, confidence, class]
predn (list): list of predictions in the native space - [xmin, ymin, xmax, ymax, confidence, class]
path (str): local path of the current evaluation image
"""
if self.val_table and self.result_table: # Log Table if Val dataset is uploaded as artifact
self.log_training_progress(predn, path, names)
if len(self.bbox_media_panel_images) < self.max_imgs_to_log and self.current_epoch > 0:
if self.current_epoch % self.bbox_interval == 0:
box_data = [{"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]},
"class_id": int(cls),
"box_caption": f"{names[cls]} {conf:.3f}",
"scores": {"class_score": conf},
"domain": "pixel"} for *xyxy, conf, cls in pred.tolist()]
boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
self.bbox_media_panel_images.append(wandb.Image(im, boxes=boxes, caption=path.name))
def log(self, log_dict):
"""
save the metrics to the logging dictionary
arguments:
log_dict (Dict) -- metrics/media to be logged in current step
"""
if self.wandb_run:
for key, value in log_dict.items():
self.log_dict[key] = value
def end_epoch(self, best_result=False):
"""
commit the log_dict, model artifacts and Tables to W&B and flush the log_dict.
arguments:
best_result (boolean): Boolean representing if the result of this evaluation is best or not
"""
if self.wandb_run:
with all_logging_disabled():
if self.bbox_media_panel_images:
self.log_dict["BoundingBoxDebugger"] = self.bbox_media_panel_images
try:
wandb.log(self.log_dict)
except BaseException as e:
LOGGER.info(f"An error occurred in wandb logger. The training will proceed without interruption. More info\n{e}")
self.wandb_run.finish()
self.wandb_run = None
self.log_dict = {}
self.bbox_media_panel_images = []
if self.result_artifact:
self.result_artifact.add(self.result_table, 'result')
wandb.log_artifact(self.result_artifact, aliases=['latest', 'last', 'epoch ' + str(self.current_epoch),
('best' if best_result else '')])
wandb.log({"evaluation": self.result_table})
self.result_table = wandb.Table(["epoch", "id", "ground truth", "prediction", "avg_confidence"])
self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
def finish_run(self):
"""
Log metrics if any and finish the current W&B run
"""
if self.wandb_run:
if self.log_dict:
with all_logging_disabled():
wandb.log(self.log_dict)
wandb.run.finish()
@contextmanager
def all_logging_disabled(highest_level=logging.CRITICAL):
""" source - https://gist.github.com/simon-weber/7853144
A context manager that will prevent any logging messages triggered during the body from being processed.
:param highest_level: the maximum logging level in use.
This would only need to be changed if a custom level greater than CRITICAL is defined.
"""
previous_level = logging.root.manager.disable
logging.disable(highest_level)
try:
yield
finally:
logging.disable(previous_level)
+24 -18
View File
@@ -1,9 +1,12 @@
# Loss functions
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Loss functions
"""
import torch
import torch.nn as nn
from utils.general import bbox_iou
from utils.metrics import bbox_iou
from utils.torch_utils import is_parallel
@@ -15,7 +18,7 @@ def smooth_BCE(eps=0.1): # https://github.com/ultralytics/yolov3/issues/238#iss
class BCEBlurWithLogitsLoss(nn.Module):
# BCEwithLogitLoss() with reduced missing label effects.
def __init__(self, alpha=0.05):
super(BCEBlurWithLogitsLoss, self).__init__()
super().__init__()
self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none') # must be nn.BCEWithLogitsLoss()
self.alpha = alpha
@@ -32,7 +35,7 @@ class BCEBlurWithLogitsLoss(nn.Module):
class FocalLoss(nn.Module):
# Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
super(FocalLoss, self).__init__()
super().__init__()
self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
self.gamma = gamma
self.alpha = alpha
@@ -62,7 +65,7 @@ class FocalLoss(nn.Module):
class QFocalLoss(nn.Module):
# Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
super(QFocalLoss, self).__init__()
super().__init__()
self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
self.gamma = gamma
self.alpha = alpha
@@ -88,7 +91,7 @@ class QFocalLoss(nn.Module):
class ComputeLoss:
# Compute losses
def __init__(self, model, autobalance=False):
super(ComputeLoss, self).__init__()
self.sort_obj_iou = False
device = next(model.parameters()).device # get model device
h = model.hyp # hyperparameters
@@ -105,9 +108,9 @@ class ComputeLoss:
BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module
self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, .02]) # P3-P7
self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, 0.02]) # P3-P7
self.ssi = list(det.stride).index(16) if autobalance else 0 # stride 16 index
self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, model.gr, h, autobalance
self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance
for k in 'na', 'nc', 'nl', 'anchors':
setattr(self, k, getattr(det, k))
@@ -126,14 +129,18 @@ class ComputeLoss:
ps = pi[b, a, gj, gi] # prediction subset corresponding to targets
# Regression
pxy = ps[:, :2].sigmoid() * 2. - 0.5
pxy = ps[:, :2].sigmoid() * 2 - 0.5
pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
pbox = torch.cat((pxy, pwh), 1) # predicted box
iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target)
lbox += (1.0 - iou).mean() # iou loss
# Objectness
tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio
score_iou = iou.detach().clamp(0).type(tobj.dtype)
if self.sort_obj_iou:
sort_id = torch.argsort(score_iou)
b, a, gj, gi, score_iou = b[sort_id], a[sort_id], gj[sort_id], gi[sort_id], score_iou[sort_id]
tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * score_iou # iou ratio
# Classification
if self.nc > 1: # cls loss (only if multiple classes)
@@ -157,8 +164,7 @@ class ComputeLoss:
lcls *= self.hyp['cls']
bs = tobj.shape[0] # batch size
loss = lbox + lobj + lcls
return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach()
return (lbox + lobj + lcls) * bs, torch.cat((lbox, lobj, lcls)).detach()
def build_targets(self, p, targets):
# Build targets for compute_loss(), input targets(image,class,x,y,w,h)
@@ -170,7 +176,7 @@ class ComputeLoss:
g = 0.5 # bias
off = torch.tensor([[0, 0],
# [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
[1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
# [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
], device=targets.device).float() * g # offsets
@@ -183,17 +189,17 @@ class ComputeLoss:
if nt:
# Matches
r = t[:, :, 4:6] / anchors[:, None] # wh ratio
j = torch.max(r, 1. / r).max(2)[0] < self.hyp['anchor_t'] # compare
j = torch.max(r, 1 / r).max(2)[0] < self.hyp['anchor_t'] # compare
# j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
t = t[j] # filter
# Offsets
gxy = t[:, 2:4] # grid xy
gxi = gain[[2, 3]] - gxy # inverse
j, k = ((gxy % 1. < g) & (gxy > 1.)).T
l, m = ((gxi % 1. < g) & (gxi > 1.)).T
j = torch.stack((torch.ones_like(j),))
t = t.repeat((off.shape[0], 1, 1))[j]
j, k = ((gxy % 1 < g) & (gxy > 1)).T
l, m = ((gxi % 1 < g) & (gxi > 1)).T
j = torch.stack((torch.ones_like(j), j, k, l, m))
t = t.repeat((5, 1, 1))[j]
offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
else:
t = targets[0]
+124 -12
View File
@@ -1,13 +1,16 @@
# Model validation metrics
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Model validation metrics
"""
import math
import warnings
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import torch
from . import general
def fitness(x):
# Model fitness as a weighted combination of metrics
@@ -68,6 +71,8 @@ def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names
# Compute F1 (harmonic mean of precision and recall)
f1 = 2 * p * r / (p + r + 1e-16)
names = [v for k, v in names.items() if k in unique_classes] # list: only classes that have data
names = {i: v for i, v in enumerate(names)} # to dict
if plot:
plot_pr_curve(px, py, ap, Path(save_dir) / 'PR_curve.png', names)
plot_mc_curve(px, f1, Path(save_dir) / 'F1_curve.png', names, ylabel='F1')
@@ -88,8 +93,8 @@ def compute_ap(recall, precision):
"""
# Append sentinel values to beginning and end
mrec = np.concatenate(([0.], recall, [recall[-1] + 0.01]))
mpre = np.concatenate(([1.], precision, [0.]))
mrec = np.concatenate(([0.0], recall, [1.0]))
mpre = np.concatenate(([1.0], precision, [0.0]))
# Compute the precision envelope
mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))
@@ -127,7 +132,7 @@ class ConfusionMatrix:
detections = detections[detections[:, 4] > self.conf]
gt_classes = labels[:, 0].int()
detection_classes = detections[:, 5].int()
iou = general.box_iou(labels[:, 1:], detections[:, :4])
iou = box_iou(labels[:, 1:], detections[:, :4])
x = torch.where(iou > self.iou_thres)
if x[0].shape[0]:
@@ -157,30 +162,135 @@ class ConfusionMatrix:
def matrix(self):
return self.matrix
def plot(self, save_dir='', names=()):
def plot(self, normalize=True, save_dir='', names=()):
try:
import seaborn as sn
array = self.matrix / (self.matrix.sum(0).reshape(1, self.nc + 1) + 1E-6) # normalize
array = self.matrix / ((self.matrix.sum(0).reshape(1, -1) + 1E-6) if normalize else 1) # normalize columns
array[array < 0.005] = np.nan # don't annotate (would appear as 0.00)
fig = plt.figure(figsize=(12, 9), tight_layout=True)
sn.set(font_scale=1.0 if self.nc < 50 else 0.8) # for label size
labels = (0 < len(names) < 99) and len(names) == self.nc # apply names to ticklabels
sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, cmap='Blues', fmt='.2f', square=True,
xticklabels=names + ['background FP'] if labels else "auto",
yticklabels=names + ['background FN'] if labels else "auto").set_facecolor((1, 1, 1))
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress empty matrix RuntimeWarning: All-NaN slice encountered
sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, cmap='Blues', fmt='.2f', square=True,
xticklabels=names + ['background FP'] if labels else "auto",
yticklabels=names + ['background FN'] if labels else "auto").set_facecolor((1, 1, 1))
fig.axes[0].set_xlabel('True')
fig.axes[0].set_ylabel('Predicted')
fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250)
plt.close()
except Exception as e:
pass
print(f'WARNING: ConfusionMatrix plot failure: {e}')
def print(self):
for i in range(self.nc + 1):
print(' '.join(map(str, self.matrix[i])))
def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7):
# Returns the IoU of box1 to box2. box1 is 4, box2 is nx4
box2 = box2.T
# Get the coordinates of bounding boxes
if x1y1x2y2: # x1, y1, x2, y2 = box1
b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
else: # transform from xywh to xyxy
b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2
b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2
b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2
b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2
# Intersection area
inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
(torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
# Union Area
w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps
union = w1 * h1 + w2 * h2 - inter + eps
iou = inter / union
if GIoU or DIoU or CIoU:
cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex (smallest enclosing box) width
ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height
if CIoU or DIoU: # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
c2 = cw ** 2 + ch ** 2 + eps # convex diagonal squared
rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 +
(b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4 # center distance squared
if DIoU:
return iou - rho2 / c2 # DIoU
elif CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
with torch.no_grad():
alpha = v / (v - iou + (1 + eps))
return iou - (rho2 / c2 + v * alpha) # CIoU
else: # GIoU https://arxiv.org/pdf/1902.09630.pdf
c_area = cw * ch + eps # convex area
return iou - (c_area - union) / c_area # GIoU
else:
return iou # IoU
def box_iou(box1, box2):
# https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
"""
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
box1 (Tensor[N, 4])
box2 (Tensor[M, 4])
Returns:
iou (Tensor[N, M]): the NxM matrix containing the pairwise
IoU values for every element in boxes1 and boxes2
"""
def box_area(box):
# box = 4xn
return (box[2] - box[0]) * (box[3] - box[1])
area1 = box_area(box1.T)
area2 = box_area(box2.T)
# inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
return inter / (area1[:, None] + area2 - inter) # iou = inter / (area1 + area2 - inter)
def bbox_ioa(box1, box2, eps=1E-7):
""" Returns the intersection over box2 area given box1, box2. Boxes are x1y1x2y2
box1: np.array of shape(4)
box2: np.array of shape(nx4)
returns: np.array of shape(n)
"""
box2 = box2.transpose()
# Get the coordinates of bounding boxes
b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
# Intersection area
inter_area = (np.minimum(b1_x2, b2_x2) - np.maximum(b1_x1, b2_x1)).clip(0) * \
(np.minimum(b1_y2, b2_y2) - np.maximum(b1_y1, b2_y1)).clip(0)
# box2 area
box2_area = (b2_x2 - b2_x1) * (b2_y2 - b2_y1) + eps
# Intersection over box2 area
return inter_area / box2_area
def wh_iou(wh1, wh2):
# Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
wh1 = wh1[:, None] # [N,1,2]
wh2 = wh2[None] # [1,M,2]
inter = torch.min(wh1, wh2).prod(2) # [N,M]
return inter / (wh1.prod(2) + wh2.prod(2) - inter) # iou = inter / (area1 + area2 - inter)
# Plots ----------------------------------------------------------------------------------------------------------------
def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()):
@@ -201,6 +311,7 @@ def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()):
ax.set_ylim(0, 1)
plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
fig.savefig(Path(save_dir), dpi=250)
plt.close()
def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence', ylabel='Metric'):
@@ -221,3 +332,4 @@ def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence'
ax.set_ylim(0, 1)
plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
fig.savefig(Path(save_dir), dpi=250)
plt.close()
+231 -208
View File
@@ -1,9 +1,10 @@
# Plotting utils
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
Plotting utils
"""
import glob
import math
import os
import random
from copy import copy
from pathlib import Path
@@ -12,15 +13,17 @@ import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
import seaborn as sn
import torch
import yaml
from PIL import Image, ImageDraw, ImageFont
from utils.general import xywh2xyxy, xyxy2xywh
from utils.general import (LOGGER, Timeout, check_requirements, clip_coords, increment_path, is_ascii, is_chinese,
try_except, user_config_dir, xywh2xyxy, xyxy2xywh)
from utils.metrics import fitness
# Settings
CONFIG_DIR = user_config_dir() # Ultralytics settings dir
RANK = int(os.getenv('RANK', -1))
matplotlib.rc('font', **{'size': 11})
matplotlib.use('Agg') # for writing to files only
@@ -46,6 +49,105 @@ class Colors:
colors = Colors() # create instance for 'from utils.plots import colors'
def check_font(font='Arial.ttf', size=10):
# Return a PIL TrueType Font, downloading to CONFIG_DIR if necessary
font = Path(font)
font = font if font.exists() else (CONFIG_DIR / font.name)
try:
return ImageFont.truetype(str(font) if font.exists() else font.name, size)
except Exception as e: # download if missing
url = "https://ultralytics.com/assets/" + font.name
print(f'Downloading {url} to {font}...')
torch.hub.download_url_to_file(url, str(font), progress=False)
try:
return ImageFont.truetype(str(font), size)
except TypeError:
check_requirements('Pillow>=8.4.0') # known issue https://github.com/ultralytics/yolov5/issues/5374
class Annotator:
if RANK in (-1, 0):
check_font() # download TTF if necessary
# Annotator for train/val mosaics and jpgs and detect/hub inference annotations
def __init__(self, im, line_width=None, font_size=None, font='Arial.ttf', pil=False, example='abc'):
assert im.data.contiguous, 'Image not contiguous. Apply np.ascontiguousarray(im) to Annotator() input images.'
self.pil = pil or not is_ascii(example) or is_chinese(example)
if self.pil: # use PIL
self.im = im if isinstance(im, Image.Image) else Image.fromarray(im)
self.draw = ImageDraw.Draw(self.im)
self.font = check_font(font='Arial.Unicode.ttf' if is_chinese(example) else font,
size=font_size or max(round(sum(self.im.size) / 2 * 0.035), 12))
else: # use cv2
self.im = im
self.lw = line_width or max(round(sum(im.shape) / 2 * 0.003), 2) # line width
def box_label(self, box, label='', color=(128, 128, 128), txt_color=(255, 255, 255)):
# Add one xyxy box to image with label
if self.pil or not is_ascii(label):
self.draw.rectangle(box, width=self.lw, outline=color) # box
if label:
w, h = self.font.getsize(label) # text width, height
outside = box[1] - h >= 0 # label fits outside box
self.draw.rectangle([box[0],
box[1] - h if outside else box[1],
box[0] + w + 1,
box[1] + 1 if outside else box[1] + h + 1], fill=color)
# self.draw.text((box[0], box[1]), label, fill=txt_color, font=self.font, anchor='ls') # for PIL>8.0
self.draw.text((box[0], box[1] - h if outside else box[1]), label, fill=txt_color, font=self.font)
else: # cv2
p1, p2 = (int(box[0]), int(box[1])), (int(box[2]), int(box[3]))
cv2.rectangle(self.im, p1, p2, color, thickness=self.lw, lineType=cv2.LINE_AA)
if label:
tf = max(self.lw - 1, 1) # font thickness
w, h = cv2.getTextSize(label, 0, fontScale=self.lw / 3, thickness=tf)[0] # text width, height
outside = p1[1] - h - 3 >= 0 # label fits outside box
p2 = p1[0] + w, p1[1] - h - 3 if outside else p1[1] + h + 3
cv2.rectangle(self.im, p1, p2, color, -1, cv2.LINE_AA) # filled
cv2.putText(self.im, label, (p1[0], p1[1] - 2 if outside else p1[1] + h + 2), 0, self.lw / 3, txt_color,
thickness=tf, lineType=cv2.LINE_AA)
def rectangle(self, xy, fill=None, outline=None, width=1):
# Add rectangle to image (PIL-only)
self.draw.rectangle(xy, fill, outline, width)
def text(self, xy, text, txt_color=(255, 255, 255)):
# Add text to image (PIL-only)
w, h = self.font.getsize(text) # text width, height
self.draw.text((xy[0], xy[1] - h + 1), text, fill=txt_color, font=self.font)
def result(self):
# Return annotated image as array
return np.asarray(self.im)
def feature_visualization(x, module_type, stage, n=32, save_dir=Path('runs/detect/exp')):
"""
x: Features to be visualized
module_type: Module type
stage: Module stage within model
n: Maximum number of feature maps to plot
save_dir: Directory to save results
"""
if 'Detect' not in module_type:
batch, channels, height, width = x.shape # batch, channels, height, width
if height > 1 and width > 1:
f = f"stage{stage}_{module_type.split('.')[-1]}_features.png" # filename
blocks = torch.chunk(x[0].cpu(), channels, dim=0) # select batch index 0, block by channels
n = min(n, channels) # number of plots
fig, ax = plt.subplots(math.ceil(n / 8), 8, tight_layout=True) # 8 rows x n/8 cols
ax = ax.ravel()
plt.subplots_adjust(wspace=0.05, hspace=0.05)
for i in range(n):
ax[i].imshow(blocks[i].squeeze()) # cmap='gray'
ax[i].axis('off')
print(f'Saving {save_dir / f}... ({n}/{channels})')
plt.savefig(save_dir / f, dpi=300, bbox_inches='tight')
plt.close()
def hist2d(x, y, n=100):
# 2d histogram used in labels.png and evolve.png
xedges, yedges = np.linspace(x.min(), x.max(), n), np.linspace(y.min(), y.max(), n)
@@ -68,54 +170,6 @@ def butter_lowpass_filtfilt(data, cutoff=1500, fs=50000, order=5):
return filtfilt(b, a, data) # forward-backward filter
def plot_one_box(x, im, color=(128, 128, 128), label=None, line_thickness=3):
# Plots one bounding box on image 'im' using OpenCV
assert im.data.contiguous, 'Image not contiguous. Apply np.ascontiguousarray(im) to plot_on_box() input image.'
tl = line_thickness or round(0.002 * (im.shape[0] + im.shape[1]) / 2) + 1 # line/font thickness
c1, c2 = (int(x[0]), int(x[1])), (int(x[2]), int(x[3]))
cv2.rectangle(im, c1, c2, color, thickness=tl, lineType=cv2.LINE_AA)
if label:
tf = max(tl - 1, 1) # font thickness
t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3
cv2.rectangle(im, c1, c2, color, -1, cv2.LINE_AA) # filled
cv2.putText(im, label, (c1[0], c1[1] - 2), 0, tl / 3, [225, 255, 255], thickness=tf, lineType=cv2.LINE_AA)
def plot_one_box_PIL(box, im, color=(128, 128, 128), label=None, line_thickness=None):
# Plots one bounding box on image 'im' using PIL
im = Image.fromarray(im)
draw = ImageDraw.Draw(im)
line_thickness = line_thickness or max(int(min(im.size) / 200), 2)
draw.rectangle(box, width=line_thickness, outline=color) # plot
if label:
font = ImageFont.truetype("Arial.ttf", size=max(round(max(im.size) / 40), 12))
txt_width, txt_height = font.getsize(label)
draw.rectangle([box[0], box[1] - txt_height + 4, box[0] + txt_width, box[1]], fill=color)
draw.text((box[0], box[1] - txt_height + 1), label, fill=(255, 255, 255), font=font)
return np.asarray(im)
def plot_wh_methods(): # from utils.plots import *; plot_wh_methods()
# Compares the two methods for width-height anchor multiplication
# https://github.com/ultralytics/yolov3/issues/168
x = np.arange(-4.0, 4.0, .1)
ya = np.exp(x)
yb = torch.sigmoid(torch.from_numpy(x)).numpy() * 2
fig = plt.figure(figsize=(6, 3), tight_layout=True)
plt.plot(x, ya, '.-', label='YOLOv3')
plt.plot(x, yb ** 2, '.-', label='YOLOv5 ^2')
plt.plot(x, yb ** 1.6, '.-', label='YOLOv5 ^1.6')
plt.xlim(left=-4, right=4)
plt.ylim(bottom=0, top=6)
plt.xlabel('input')
plt.ylabel('output')
plt.grid()
plt.legend()
fig.savefig('comparison.png', dpi=200)
def output_to_target(output):
# Convert model output to target format [batch_id, class_id, x, y, w, h, conf]
targets = []
@@ -125,82 +179,65 @@ def output_to_target(output):
return np.array(targets)
def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=640, max_subplots=16):
def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=1920, max_subplots=16):
# Plot image grid with labels
if isinstance(images, torch.Tensor):
images = images.cpu().float().numpy()
if isinstance(targets, torch.Tensor):
targets = targets.cpu().numpy()
# un-normalise
if np.max(images[0]) <= 1:
images *= 255
tl = 3 # line thickness
tf = max(tl - 1, 1) # font thickness
images *= 255 # de-normalise (optional)
bs, _, h, w = images.shape # batch size, _, height, width
bs = min(bs, max_subplots) # limit plot images
ns = np.ceil(bs ** 0.5) # number of subplots (square)
# Check if we should resize
scale_factor = max_size / max(h, w)
if scale_factor < 1:
h = math.ceil(scale_factor * h)
w = math.ceil(scale_factor * w)
# Build Image
mosaic = np.full((int(ns * h), int(ns * w), 3), 255, dtype=np.uint8) # init
for i, img in enumerate(images):
for i, im in enumerate(images):
if i == max_subplots: # if last batch has fewer images than we expect
break
x, y = int(w * (i // ns)), int(h * (i % ns)) # block origin
im = im.transpose(1, 2, 0)
mosaic[y:y + h, x:x + w, :] = im
block_x = int(w * (i // ns))
block_y = int(h * (i % ns))
# Resize (optional)
scale = max_size / ns / max(h, w)
if scale < 1:
h = math.ceil(scale * h)
w = math.ceil(scale * w)
mosaic = cv2.resize(mosaic, tuple(int(x * ns) for x in (w, h)))
img = img.transpose(1, 2, 0)
if scale_factor < 1:
img = cv2.resize(img, (w, h))
mosaic[block_y:block_y + h, block_x:block_x + w, :] = img
# Annotate
fs = int((h + w) * ns * 0.01) # font size
annotator = Annotator(mosaic, line_width=round(fs / 10), font_size=fs, pil=True)
for i in range(i + 1):
x, y = int(w * (i // ns)), int(h * (i % ns)) # block origin
annotator.rectangle([x, y, x + w, y + h], None, (255, 255, 255), width=2) # borders
if paths:
annotator.text((x + 5, y + 5 + h), text=Path(paths[i]).name[:40], txt_color=(220, 220, 220)) # filenames
if len(targets) > 0:
image_targets = targets[targets[:, 0] == i]
boxes = xywh2xyxy(image_targets[:, 2:6]).T
classes = image_targets[:, 1].astype('int')
labels = image_targets.shape[1] == 6 # labels if no conf column
conf = None if labels else image_targets[:, 6] # check for confidence presence (label vs pred)
ti = targets[targets[:, 0] == i] # image targets
boxes = xywh2xyxy(ti[:, 2:6]).T
classes = ti[:, 1].astype('int')
labels = ti.shape[1] == 6 # labels if no conf column
conf = None if labels else ti[:, 6] # check for confidence presence (label vs pred)
if boxes.shape[1]:
if boxes.max() <= 1.01: # if normalized with tolerance 0.01
boxes[[0, 2]] *= w # scale to pixels
boxes[[1, 3]] *= h
elif scale_factor < 1: # absolute coords need scale if image scales
boxes *= scale_factor
boxes[[0, 2]] += block_x
boxes[[1, 3]] += block_y
for j, box in enumerate(boxes.T):
cls = int(classes[j])
elif scale < 1: # absolute coords need scale if image scales
boxes *= scale
boxes[[0, 2]] += x
boxes[[1, 3]] += y
for j, box in enumerate(boxes.T.tolist()):
cls = classes[j]
color = colors(cls)
cls = names[cls] if names else cls
if labels or conf[j] > 0.25: # 0.25 conf thresh
label = '%s' % cls if labels else '%s %.1f' % (cls, conf[j])
plot_one_box(box, mosaic, label=label, color=color, line_thickness=tl)
# Draw image filename labels
if paths:
label = Path(paths[i]).name[:40] # trim to 40 char
t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
cv2.putText(mosaic, label, (block_x + 5, block_y + t_size[1] + 5), 0, tl / 3, [220, 220, 220], thickness=tf,
lineType=cv2.LINE_AA)
# Image border
cv2.rectangle(mosaic, (block_x, block_y), (block_x + w, block_y + h), (255, 255, 255), thickness=3)
if fname:
r = min(1280. / max(h, w) / ns, 1.0) # ratio to limit image size
mosaic = cv2.resize(mosaic, (int(ns * w * r), int(ns * h * r)), interpolation=cv2.INTER_AREA)
# cv2.imwrite(fname, cv2.cvtColor(mosaic, cv2.COLOR_BGR2RGB)) # cv2 save
Image.fromarray(mosaic).save(fname) # PIL save
return mosaic
label = f'{cls}' if labels else f'{cls} {conf[j]:.1f}'
annotator.box_label(box, label, color=color)
annotator.im.save(fname) # save
def plot_lr_scheduler(optimizer, scheduler, epochs=300, save_dir=''):
@@ -220,9 +257,9 @@ def plot_lr_scheduler(optimizer, scheduler, epochs=300, save_dir=''):
plt.close()
def plot_test_txt(): # from utils.plots import *; plot_test()
# Plot test.txt histograms
x = np.loadtxt('test.txt', dtype=np.float32)
def plot_val_txt(): # from utils.plots import *; plot_val()
# Plot val.txt histograms
x = np.loadtxt('val.txt', dtype=np.float32)
box = xyxy2xywh(x[:, :4])
cx, cy = box[:, 0], box[:, 1]
@@ -244,29 +281,32 @@ def plot_targets_txt(): # from utils.plots import *; plot_targets_txt()
fig, ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)
ax = ax.ravel()
for i in range(4):
ax[i].hist(x[i], bins=100, label='%.3g +/- %.3g' % (x[i].mean(), x[i].std()))
ax[i].hist(x[i], bins=100, label=f'{x[i].mean():.3g} +/- {x[i].std():.3g}')
ax[i].legend()
ax[i].set_title(s[i])
plt.savefig('targets.jpg', dpi=200)
def plot_study_txt(path='', x=None): # from utils.plots import *; plot_study_txt()
# Plot study.txt generated by test.py
fig, ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)
# ax = ax.ravel()
def plot_val_study(file='', dir='', x=None): # from utils.plots import *; plot_val_study()
# Plot file=study.txt generated by val.py (or plot all study*.txt in dir)
save_dir = Path(file).parent if file else Path(dir)
plot2 = False # plot additional results
if plot2:
ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)[1].ravel()
fig2, ax2 = plt.subplots(1, 1, figsize=(8, 4), tight_layout=True)
# for f in [Path(path) / f'study_coco_{x}.txt' for x in ['yolov3-tiny', 'yolov3', 'yolov3-spp', 'yolov5l']]:
for f in sorted(Path(path).glob('study*.txt')):
# for f in [save_dir / f'study_coco_{x}.txt' for x in ['yolov3', 'yolov3-spp', 'yolov3-tiny']]:
for f in sorted(save_dir.glob('study*.txt')):
y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
x = np.arange(y.shape[1]) if x is None else np.array(x)
s = ['P', 'R', 'mAP@.5', 'mAP@.5:.95', 't_inference (ms/img)', 't_NMS (ms/img)', 't_total (ms/img)']
# for i in range(7):
# ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
# ax[i].set_title(s[i])
if plot2:
s = ['P', 'R', 'mAP@.5', 'mAP@.5:.95', 't_preprocess (ms/img)', 't_inference (ms/img)', 't_NMS (ms/img)']
for i in range(7):
ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
ax[i].set_title(s[i])
j = y[3].argmax() + 1
ax2.plot(y[6, 1:j], y[3, 1:j] * 1E2, '.-', linewidth=2, markersize=8,
ax2.plot(y[5, 1:j], y[3, 1:j] * 1E2, '.-', linewidth=2, markersize=8,
label=f.stem.replace('study_coco_', '').replace('yolo', 'YOLO'))
ax2.plot(1E3 / np.array([209, 140, 97, 58, 35, 18]), [34.6, 40.5, 43.0, 47.5, 49.7, 51.5],
@@ -275,22 +315,26 @@ def plot_study_txt(path='', x=None): # from utils.plots import *; plot_study_tx
ax2.grid(alpha=0.2)
ax2.set_yticks(np.arange(20, 60, 5))
ax2.set_xlim(0, 57)
ax2.set_ylim(15, 55)
ax2.set_ylim(25, 55)
ax2.set_xlabel('GPU Speed (ms/img)')
ax2.set_ylabel('COCO AP val')
ax2.legend(loc='lower right')
plt.savefig(str(Path(path).name) + '.png', dpi=300)
f = save_dir / 'study.png'
print(f'Saving {f}...')
plt.savefig(f, dpi=300)
def plot_labels(labels, names=(), save_dir=Path(''), loggers=None):
@try_except # known issue https://github.com/ultralytics/yolov5/issues/5395
@Timeout(30) # known issue https://github.com/ultralytics/yolov5/issues/5611
def plot_labels(labels, names=(), save_dir=Path('')):
# plot dataset labels
print('Plotting labels... ')
LOGGER.info(f"Plotting labels to {save_dir / 'labels.jpg'}... ")
c, b = labels[:, 0], labels[:, 1:].transpose() # classes, boxes
nc = int(c.max() + 1) # number of classes
x = pd.DataFrame(b.transpose(), columns=['x', 'y', 'width', 'height'])
# seaborn correlogram
sns.pairplot(x, corner=True, diag_kind='auto', kind='hist', diag_kws=dict(bins=50), plot_kws=dict(pmax=0.9))
sn.pairplot(x, corner=True, diag_kind='auto', kind='hist', diag_kws=dict(bins=50), plot_kws=dict(pmax=0.9))
plt.savefig(save_dir / 'labels_correlogram.jpg', dpi=200)
plt.close()
@@ -298,15 +342,15 @@ def plot_labels(labels, names=(), save_dir=Path(''), loggers=None):
matplotlib.use('svg') # faster
ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)[1].ravel()
y = ax[0].hist(c, bins=np.linspace(0, nc, nc + 1) - 0.5, rwidth=0.8)
# [y[2].patches[i].set_color([x / 255 for x in colors(i)]) for i in range(nc)] # update colors bug #3195
# [y[2].patches[i].set_color([x / 255 for x in colors(i)]) for i in range(nc)] # update colors bug #3195
ax[0].set_ylabel('instances')
if 0 < len(names) < 30:
ax[0].set_xticks(range(len(names)))
ax[0].set_xticklabels(names, rotation=90, fontsize=10)
else:
ax[0].set_xlabel('classes')
sns.histplot(x, x='x', y='y', ax=ax[2], bins=50, pmax=0.9)
sns.histplot(x, x='width', y='height', ax=ax[3], bins=50, pmax=0.9)
sn.histplot(x, x='x', y='y', ax=ax[2], bins=50, pmax=0.9)
sn.histplot(x, x='width', y='height', ax=ax[3], bins=50, pmax=0.9)
# rectangles
labels[:, 1:3] = 0.5 # center
@@ -325,34 +369,57 @@ def plot_labels(labels, names=(), save_dir=Path(''), loggers=None):
matplotlib.use('Agg')
plt.close()
# loggers
for k, v in loggers.items() or {}:
if k == 'wandb' and v:
v.log({"Labels": [v.Image(str(x), caption=x.name) for x in save_dir.glob('*labels*.jpg')]}, commit=False)
def plot_evolution(yaml_file='data/hyp.finetune.yaml'): # from utils.plots import *; plot_evolution()
# Plot hyperparameter evolution results in evolve.txt
with open(yaml_file) as f:
hyp = yaml.safe_load(f)
x = np.loadtxt('evolve.txt', ndmin=2)
def plot_evolve(evolve_csv='path/to/evolve.csv'): # from utils.plots import *; plot_evolve()
# Plot evolve.csv hyp evolution results
evolve_csv = Path(evolve_csv)
data = pd.read_csv(evolve_csv)
keys = [x.strip() for x in data.columns]
x = data.values
f = fitness(x)
# weights = (f - f.min()) ** 2 # for weighted results
j = np.argmax(f) # max fitness index
plt.figure(figsize=(10, 12), tight_layout=True)
matplotlib.rc('font', **{'size': 8})
for i, (k, v) in enumerate(hyp.items()):
y = x[:, i + 7]
# mu = (y * weights).sum() / weights.sum() # best weighted result
mu = y[f.argmax()] # best single result
for i, k in enumerate(keys[7:]):
v = x[:, 7 + i]
mu = v[j] # best single result
plt.subplot(6, 5, i + 1)
plt.scatter(y, f, c=hist2d(y, f, 20), cmap='viridis', alpha=.8, edgecolors='none')
plt.scatter(v, f, c=hist2d(v, f, 20), cmap='viridis', alpha=.8, edgecolors='none')
plt.plot(mu, f.max(), 'k+', markersize=15)
plt.title('%s = %.3g' % (k, mu), fontdict={'size': 9}) # limit to 40 characters
plt.title(f'{k} = {mu:.3g}', fontdict={'size': 9}) # limit to 40 characters
if i % 5 != 0:
plt.yticks([])
print('%15s: %.3g' % (k, mu))
plt.savefig('evolve.png', dpi=200)
print('\nPlot saved as evolve.png')
print(f'{k:>15}: {mu:.3g}')
f = evolve_csv.with_suffix('.png') # filename
plt.savefig(f, dpi=200)
plt.close()
print(f'Saved {f}')
def plot_results(file='path/to/results.csv', dir=''):
# Plot training results.csv. Usage: from utils.plots import *; plot_results('path/to/results.csv')
save_dir = Path(file).parent if file else Path(dir)
fig, ax = plt.subplots(2, 5, figsize=(12, 6), tight_layout=True)
ax = ax.ravel()
files = list(save_dir.glob('results*.csv'))
assert len(files), f'No results.csv files found in {save_dir.resolve()}, nothing to plot.'
for fi, f in enumerate(files):
try:
data = pd.read_csv(f)
s = [x.strip() for x in data.columns]
x = data.values[:, 0]
for i, j in enumerate([1, 2, 3, 4, 5, 8, 9, 10, 6, 7]):
y = data.values[:, j]
# y[y == 0] = np.nan # don't show zero values
ax[i].plot(x, y, marker='.', label=f.stem, linewidth=2, markersize=8)
ax[i].set_title(s[j], fontsize=12)
# if j in [8, 9, 10]: # share train and val loss y axes
# ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])
except Exception as e:
print(f'Warning: Plotting error for {f}: {e}')
ax[1].legend()
fig.savefig(save_dir / 'results.png', dpi=200)
plt.close()
def profile_idetection(start=0, stop=0, labels=(), save_dir=''):
@@ -381,66 +448,22 @@ def profile_idetection(start=0, stop=0, labels=(), save_dir=''):
else:
a.remove()
except Exception as e:
print('Warning: Plotting error for %s; %s' % (f, e))
print(f'Warning: Plotting error for {f}; {e}')
ax[1].legend()
plt.savefig(Path(save_dir) / 'idetection_profile.png', dpi=200)
def plot_results_overlay(start=0, stop=0): # from utils.plots import *; plot_results_overlay()
# Plot training 'results*.txt', overlaying train and val losses
s = ['train', 'train', 'train', 'Precision', 'mAP@0.5', 'val', 'val', 'val', 'Recall', 'mAP@0.5:0.95'] # legends
t = ['Box', 'Objectness', 'Classification', 'P-R', 'mAP-F1'] # titles
for f in sorted(glob.glob('results*.txt') + glob.glob('../../Downloads/results*.txt')):
results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
n = results.shape[1] # number of rows
x = range(start, min(stop, n) if stop else n)
fig, ax = plt.subplots(1, 5, figsize=(14, 3.5), tight_layout=True)
ax = ax.ravel()
for i in range(5):
for j in [i, i + 5]:
y = results[j, x]
ax[i].plot(x, y, marker='.', label=s[j])
# y_smooth = butter_lowpass_filtfilt(y)
# ax[i].plot(x, np.gradient(y_smooth), marker='.', label=s[j])
ax[i].set_title(t[i])
ax[i].legend()
ax[i].set_ylabel(f) if i == 0 else None # add filename
fig.savefig(f.replace('.txt', '.png'), dpi=200)
def plot_results(start=0, stop=0, bucket='', id=(), labels=(), save_dir=''):
# Plot training 'results*.txt'. from utils.plots import *; plot_results(save_dir='runs/train/exp')
fig, ax = plt.subplots(2, 5, figsize=(12, 6), tight_layout=True)
ax = ax.ravel()
s = ['Box', 'Objectness', 'Classification', 'Precision', 'Recall',
'val Box', 'val Objectness', 'val Classification', 'mAP@0.5', 'mAP@0.5:0.95']
if bucket:
# files = ['https://storage.googleapis.com/%s/results%g.txt' % (bucket, x) for x in id]
files = ['results%g.txt' % x for x in id]
c = ('gsutil cp ' + '%s ' * len(files) + '.') % tuple('gs://%s/results%g.txt' % (bucket, x) for x in id)
os.system(c)
else:
files = list(Path(save_dir).glob('results*.txt'))
assert len(files), 'No results.txt files found in %s, nothing to plot.' % os.path.abspath(save_dir)
for fi, f in enumerate(files):
try:
results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
n = results.shape[1] # number of rows
x = range(start, min(stop, n) if stop else n)
for i in range(10):
y = results[i, x]
if i in [0, 1, 2, 5, 6, 7]:
y[y == 0] = np.nan # don't show zero loss values
# y /= y[0] # normalize
label = labels[fi] if len(labels) else f.stem
ax[i].plot(x, y, marker='.', label=label, linewidth=2, markersize=8)
ax[i].set_title(s[i])
# if i in [5, 6, 7]: # share train and val loss y axes
# ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])
except Exception as e:
print('Warning: Plotting error for %s; %s' % (f, e))
ax[1].legend()
fig.savefig(Path(save_dir) / 'results.png', dpi=200)
def save_one_box(xyxy, im, file='image.jpg', gain=1.02, pad=10, square=False, BGR=False, save=True):
# Save image crop as {file} with crop size multiple {gain} and {pad} pixels. Save and/or return crop
xyxy = torch.tensor(xyxy).view(-1, 4)
b = xyxy2xywh(xyxy) # boxes
if square:
b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1) # attempt rectangle to square
b[:, 2:] = b[:, 2:] * gain + pad # box wh * gain + pad
xyxy = xywh2xyxy(b).long()
clip_coords(xyxy, im.shape)
crop = im[int(xyxy[0, 1]):int(xyxy[0, 3]), int(xyxy[0, 0]):int(xyxy[0, 2]), ::(1 if BGR else -1)]
if save:
file.parent.mkdir(parents=True, exist_ok=True) # make directory
cv2.imwrite(str(increment_path(file).with_suffix('.jpg')), crop)
return crop
+96 -88
View File
@@ -1,7 +1,9 @@
# YOLOv3 PyTorch utils
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
PyTorch utils
"""
import datetime
import logging
import math
import os
import platform
@@ -12,16 +14,16 @@ from copy import deepcopy
from pathlib import Path
import torch
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.nn as nn
import torch.nn.functional as F
import torchvision
from utils.general import LOGGER
try:
import thop # for FLOPS computation
import thop # for FLOPs computation
except ImportError:
thop = None
logger = logging.getLogger(__name__)
@contextmanager
@@ -30,19 +32,10 @@ def torch_distributed_zero_first(local_rank: int):
Decorator to make all processes in distributed training wait for each local_master to do something.
"""
if local_rank not in [-1, 0]:
torch.distributed.barrier()
dist.barrier(device_ids=[local_rank])
yield
if local_rank == 0:
torch.distributed.barrier()
def init_torch_seeds(seed=0):
# Speed-reproducibility tradeoff https://pytorch.org/docs/stable/notes/randomness.html
torch.manual_seed(seed)
if seed == 0: # slower, more reproducible
cudnn.benchmark, cudnn.deterministic = False, True
else: # faster, less reproducible
cudnn.benchmark, cudnn.deterministic = True, False
dist.barrier(device_ids=[0])
def date_modified(path=__file__):
@@ -60,10 +53,11 @@ def git_describe(path=Path(__file__).parent): # path must be a directory
return '' # not a git repository
def select_device(device='', batch_size=None):
def select_device(device='', batch_size=None, newline=True):
# device = 'cpu' or '0' or '0,1,2,3'
s = f'YOLOv3 🚀 {git_describe() or date_modified()} torch {torch.__version__} ' # string
cpu = device.lower() == 'cpu'
device = str(device).strip().lower().replace('cuda:', '') # to string, 'cuda:0' to '0'
cpu = device == 'cpu'
if cpu:
os.environ['CUDA_VISIBLE_DEVICES'] = '-1' # force torch.cuda.is_available() = False
elif device: # non-cpu device requested
@@ -72,65 +66,80 @@ def select_device(device='', batch_size=None):
cuda = not cpu and torch.cuda.is_available()
if cuda:
devices = device.split(',') if device else range(torch.cuda.device_count()) # i.e. 0,1,6,7
devices = device.split(',') if device else '0' # range(torch.cuda.device_count()) # i.e. 0,1,6,7
n = len(devices) # device count
if n > 1 and batch_size: # check batch_size is divisible by device_count
assert batch_size % n == 0, f'batch-size {batch_size} not multiple of GPU count {n}'
space = ' ' * len(s)
space = ' ' * (len(s) + 1)
for i, d in enumerate(devices):
p = torch.cuda.get_device_properties(i)
s += f"{'' if i == 0 else space}CUDA:{d} ({p.name}, {p.total_memory / 1024 ** 2}MB)\n" # bytes to MB
s += f"{'' if i == 0 else space}CUDA:{d} ({p.name}, {p.total_memory / 1024 ** 2:.0f}MiB)\n" # bytes to MB
else:
s += 'CPU\n'
logger.info(s.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else s) # emoji-safe
if not newline:
s = s.rstrip()
LOGGER.info(s.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else s) # emoji-safe
return torch.device('cuda:0' if cuda else 'cpu')
def time_synchronized():
def time_sync():
# pytorch-accurate time
if torch.cuda.is_available():
torch.cuda.synchronize()
return time.time()
def profile(x, ops, n=100, device=None):
# profile a pytorch module or list of modules. Example usage:
# x = torch.randn(16, 3, 640, 640) # input
def profile(input, ops, n=10, device=None):
# speed/memory/FLOPs profiler
#
# Usage:
# input = torch.randn(16, 3, 640, 640)
# m1 = lambda x: x * torch.sigmoid(x)
# m2 = nn.SiLU()
# profile(x, [m1, m2], n=100) # profile speed over 100 iterations
# profile(input, [m1, m2], n=100) # profile over 100 iterations
device = device or torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
x = x.to(device)
x.requires_grad = True
print(torch.__version__, device.type, torch.cuda.get_device_properties(0) if device.type == 'cuda' else '')
print(f"\n{'Params':>12s}{'GFLOPS':>12s}{'forward (ms)':>16s}{'backward (ms)':>16s}{'input':>24s}{'output':>24s}")
for m in ops if isinstance(ops, list) else [ops]:
m = m.to(device) if hasattr(m, 'to') else m # device
m = m.half() if hasattr(m, 'half') and isinstance(x, torch.Tensor) and x.dtype is torch.float16 else m # type
dtf, dtb, t = 0., 0., [0., 0., 0.] # dt forward, backward
try:
flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # GFLOPS
except:
flops = 0
results = []
device = device or select_device()
print(f"{'Params':>12s}{'GFLOPs':>12s}{'GPU_mem (GB)':>14s}{'forward (ms)':>14s}{'backward (ms)':>14s}"
f"{'input':>24s}{'output':>24s}")
for _ in range(n):
t[0] = time_synchronized()
y = m(x)
t[1] = time_synchronized()
for x in input if isinstance(input, list) else [input]:
x = x.to(device)
x.requires_grad = True
for m in ops if isinstance(ops, list) else [ops]:
m = m.to(device) if hasattr(m, 'to') else m # device
m = m.half() if hasattr(m, 'half') and isinstance(x, torch.Tensor) and x.dtype is torch.float16 else m
tf, tb, t = 0, 0, [0, 0, 0] # dt forward, backward
try:
_ = y.sum().backward()
t[2] = time_synchronized()
except: # no backward method
t[2] = float('nan')
dtf += (t[1] - t[0]) * 1000 / n # ms per op forward
dtb += (t[2] - t[1]) * 1000 / n # ms per op backward
flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # GFLOPs
except:
flops = 0
s_in = tuple(x.shape) if isinstance(x, torch.Tensor) else 'list'
s_out = tuple(y.shape) if isinstance(y, torch.Tensor) else 'list'
p = sum(list(x.numel() for x in m.parameters())) if isinstance(m, nn.Module) else 0 # parameters
print(f'{p:12}{flops:12.4g}{dtf:16.4g}{dtb:16.4g}{str(s_in):>24s}{str(s_out):>24s}')
try:
for _ in range(n):
t[0] = time_sync()
y = m(x)
t[1] = time_sync()
try:
_ = (sum(yi.sum() for yi in y) if isinstance(y, list) else y).sum().backward()
t[2] = time_sync()
except Exception as e: # no backward method
# print(e) # for debug
t[2] = float('nan')
tf += (t[1] - t[0]) * 1000 / n # ms per op forward
tb += (t[2] - t[1]) * 1000 / n # ms per op backward
mem = torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0 # (GB)
s_in = tuple(x.shape) if isinstance(x, torch.Tensor) else 'list'
s_out = tuple(y.shape) if isinstance(y, torch.Tensor) else 'list'
p = sum(list(x.numel() for x in m.parameters())) if isinstance(m, nn.Module) else 0 # parameters
print(f'{p:12}{flops:12.4g}{mem:>14.3f}{tf:14.4g}{tb:14.4g}{str(s_in):>24s}{str(s_out):>24s}')
results.append([p, flops, mem, tf, tb, s_in, s_out])
except Exception as e:
print(e)
results.append(None)
torch.cuda.empty_cache()
return results
def is_parallel(model):
@@ -143,11 +152,6 @@ def de_parallel(model):
return model.module if is_parallel(model) else model
def intersect_dicts(da, db, exclude=()):
# Dictionary intersection of matching keys and shapes, omitting 'exclude' keys, using da values
return {k: v for k, v in da.items() if k in db and not any(x in k for x in exclude) and v.shape == db[k].shape}
def initialize_weights(model):
for m in model.modules():
t = type(m)
@@ -156,7 +160,7 @@ def initialize_weights(model):
elif t is nn.BatchNorm2d:
m.eps = 1e-3
m.momentum = 0.03
elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6]:
elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
m.inplace = True
@@ -167,7 +171,7 @@ def find_modules(model, mclass=nn.Conv2d):
def sparsity(model):
# Return global model sparsity
a, b = 0., 0.
a, b = 0, 0
for p in model.parameters():
a += p.numel()
b += (p == 0).sum()
@@ -213,42 +217,23 @@ def model_info(model, verbose=False, img_size=640):
n_p = sum(x.numel() for x in model.parameters()) # number parameters
n_g = sum(x.numel() for x in model.parameters() if x.requires_grad) # number gradients
if verbose:
print('%5s %40s %9s %12s %20s %10s %10s' % ('layer', 'name', 'gradient', 'parameters', 'shape', 'mu', 'sigma'))
print(f"{'layer':>5} {'name':>40} {'gradient':>9} {'parameters':>12} {'shape':>20} {'mu':>10} {'sigma':>10}")
for i, (name, p) in enumerate(model.named_parameters()):
name = name.replace('module_list.', '')
print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
(i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
try: # FLOPS
try: # FLOPs
from thop import profile
stride = max(int(model.stride.max()), 32) if hasattr(model, 'stride') else 32
img = torch.zeros((1, model.yaml.get('ch', 3), stride, stride), device=next(model.parameters()).device) # input
flops = profile(deepcopy(model), inputs=(img,), verbose=False)[0] / 1E9 * 2 # stride GFLOPS
flops = profile(deepcopy(model), inputs=(img,), verbose=False)[0] / 1E9 * 2 # stride GFLOPs
img_size = img_size if isinstance(img_size, list) else [img_size, img_size] # expand if int/float
fs = ', %.1f GFLOPS' % (flops * img_size[0] / stride * img_size[1] / stride) # 640x640 GFLOPS
fs = ', %.1f GFLOPs' % (flops * img_size[0] / stride * img_size[1] / stride) # 640x640 GFLOPs
except (ImportError, Exception):
fs = ''
logger.info(f"Model Summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}")
def load_classifier(name='resnet101', n=2):
# Loads a pretrained model reshaped to n-class output
model = torchvision.models.__dict__[name](pretrained=True)
# ResNet model properties
# input_size = [3, 224, 224]
# input_space = 'RGB'
# input_range = [0, 1]
# mean = [0.485, 0.456, 0.406]
# std = [0.229, 0.224, 0.225]
# Reshape output to n classes
filters = model.fc.weight.shape[1]
model.fc.bias = nn.Parameter(torch.zeros(n), requires_grad=True)
model.fc.weight = nn.Parameter(torch.zeros(n, filters), requires_grad=True)
model.fc.out_features = n
return model
LOGGER.info(f"Model Summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}")
def scale_img(img, ratio=1.0, same_shape=False, gs=32): # img(16,3,256,416)
@@ -260,7 +245,7 @@ def scale_img(img, ratio=1.0, same_shape=False, gs=32): # img(16,3,256,416)
s = (int(h * ratio), int(w * ratio)) # new size
img = F.interpolate(img, size=s, mode='bilinear', align_corners=False) # resize
if not same_shape: # pad/crop img
h, w = [math.ceil(x * ratio / gs) * gs for x in (h, w)]
h, w = (math.ceil(x * ratio / gs) * gs for x in (h, w))
return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447) # value = imagenet mean
@@ -273,6 +258,29 @@ def copy_attr(a, b, include=(), exclude=()):
setattr(a, k, v)
class EarlyStopping:
# simple early stopper
def __init__(self, patience=30):
self.best_fitness = 0.0 # i.e. mAP
self.best_epoch = 0
self.patience = patience or float('inf') # epochs to wait after fitness stops improving to stop
self.possible_stop = False # possible stop may occur next epoch
def __call__(self, epoch, fitness):
if fitness >= self.best_fitness: # >= 0 to allow for early zero-fitness stage of training
self.best_epoch = epoch
self.best_fitness = fitness
delta = epoch - self.best_epoch # epochs without improvement
self.possible_stop = delta >= (self.patience - 1) # possible stop may occur next epoch
stop = delta >= self.patience # stop training if patience exceeded
if stop:
LOGGER.info(f'Stopping training early as no improvement observed in last {self.patience} epochs. '
f'Best results observed at epoch {self.best_epoch}, best model saved as best.pt.\n'
f'To update EarlyStopping(patience={self.patience}) pass a new patience value, '
f'i.e. `python train.py --patience 300` or use `--patience 0` to disable EarlyStopping.')
return stop
class ModelEMA:
""" Model Exponential Moving Average from https://github.com/rwightman/pytorch-image-models
Keep a moving average of everything in the model state_dict (parameters and buffers).
@@ -303,7 +311,7 @@ class ModelEMA:
for k, v in self.ema.state_dict().items():
if v.dtype.is_floating_point:
v *= d
v += (1. - d) * msd[k].detach()
v += (1 - d) * msd[k].detach()
def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
# Update EMA attributes
View File
-318
View File
@@ -1,318 +0,0 @@
"""Utilities and tools for tracking runs with Weights & Biases."""
import json
import sys
from pathlib import Path
import torch
import yaml
from tqdm import tqdm
sys.path.append(str(Path(__file__).parent.parent.parent)) # add utils/ to path
from utils.datasets import LoadImagesAndLabels
from utils.datasets import img2label_paths
from utils.general import colorstr, xywh2xyxy, check_dataset, check_file
try:
import wandb
from wandb import init, finish
except ImportError:
wandb = None
WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
def remove_prefix(from_string, prefix=WANDB_ARTIFACT_PREFIX):
return from_string[len(prefix):]
def check_wandb_config_file(data_config_file):
wandb_config = '_wandb.'.join(data_config_file.rsplit('.', 1)) # updated data.yaml path
if Path(wandb_config).is_file():
return wandb_config
return data_config_file
def get_run_info(run_path):
run_path = Path(remove_prefix(run_path, WANDB_ARTIFACT_PREFIX))
run_id = run_path.stem
project = run_path.parent.stem
entity = run_path.parent.parent.stem
model_artifact_name = 'run_' + run_id + '_model'
return entity, project, run_id, model_artifact_name
def check_wandb_resume(opt):
process_wandb_config_ddp_mode(opt) if opt.global_rank not in [-1, 0] else None
if isinstance(opt.resume, str):
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
if opt.global_rank not in [-1, 0]: # For resuming DDP runs
entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
api = wandb.Api()
artifact = api.artifact(entity + '/' + project + '/' + model_artifact_name + ':latest')
modeldir = artifact.download()
opt.weights = str(Path(modeldir) / "last.pt")
return True
return None
def process_wandb_config_ddp_mode(opt):
with open(check_file(opt.data)) as f:
data_dict = yaml.safe_load(f) # data dict
train_dir, val_dir = None, None
if isinstance(data_dict['train'], str) and data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX):
api = wandb.Api()
train_artifact = api.artifact(remove_prefix(data_dict['train']) + ':' + opt.artifact_alias)
train_dir = train_artifact.download()
train_path = Path(train_dir) / 'data/images/'
data_dict['train'] = str(train_path)
if isinstance(data_dict['val'], str) and data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX):
api = wandb.Api()
val_artifact = api.artifact(remove_prefix(data_dict['val']) + ':' + opt.artifact_alias)
val_dir = val_artifact.download()
val_path = Path(val_dir) / 'data/images/'
data_dict['val'] = str(val_path)
if train_dir or val_dir:
ddp_data_path = str(Path(val_dir) / 'wandb_local_data.yaml')
with open(ddp_data_path, 'w') as f:
yaml.safe_dump(data_dict, f)
opt.data = ddp_data_path
class WandbLogger():
"""Log training runs, datasets, models, and predictions to Weights & Biases.
This logger sends information to W&B at wandb.ai. By default, this information
includes hyperparameters, system configuration and metrics, model metrics,
and basic data metrics and analyses.
By providing additional command line arguments to train.py, datasets,
models and predictions can also be logged.
For more on how this logger is used, see the Weights & Biases documentation:
https://docs.wandb.com/guides/integrations/yolov5
"""
def __init__(self, opt, name, run_id, data_dict, job_type='Training'):
# Pre-training routine --
self.job_type = job_type
self.wandb, self.wandb_run, self.data_dict = wandb, None if not wandb else wandb.run, data_dict
# It's more elegant to stick to 1 wandb.init call, but useful config data is overwritten in the WandbLogger's wandb.init call
if isinstance(opt.resume, str): # checks resume from artifact
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
model_artifact_name = WANDB_ARTIFACT_PREFIX + model_artifact_name
assert wandb, 'install wandb to resume wandb runs'
# Resume wandb-artifact:// runs here| workaround for not overwriting wandb.config
self.wandb_run = wandb.init(id=run_id, project=project, entity=entity, resume='allow')
opt.resume = model_artifact_name
elif self.wandb:
self.wandb_run = wandb.init(config=opt,
resume="allow",
project='YOLOv3' if opt.project == 'runs/train' else Path(opt.project).stem,
entity=opt.entity,
name=name,
job_type=job_type,
id=run_id) if not wandb.run else wandb.run
if self.wandb_run:
if self.job_type == 'Training':
if not opt.resume:
wandb_data_dict = self.check_and_upload_dataset(opt) if opt.upload_dataset else data_dict
# Info useful for resuming from artifacts
self.wandb_run.config.opt = vars(opt)
self.wandb_run.config.data_dict = wandb_data_dict
self.data_dict = self.setup_training(opt, data_dict)
if self.job_type == 'Dataset Creation':
self.data_dict = self.check_and_upload_dataset(opt)
else:
prefix = colorstr('wandb: ')
print(f"{prefix}Install Weights & Biases for YOLOv3 logging with 'pip install wandb' (recommended)")
def check_and_upload_dataset(self, opt):
assert wandb, 'Install wandb to upload dataset'
check_dataset(self.data_dict)
config_path = self.log_dataset_artifact(check_file(opt.data),
opt.single_cls,
'YOLOv3' if opt.project == 'runs/train' else Path(opt.project).stem)
print("Created dataset config file ", config_path)
with open(config_path) as f:
wandb_data_dict = yaml.safe_load(f)
return wandb_data_dict
def setup_training(self, opt, data_dict):
self.log_dict, self.current_epoch, self.log_imgs = {}, 0, 16 # Logging Constants
self.bbox_interval = opt.bbox_interval
if isinstance(opt.resume, str):
modeldir, _ = self.download_model_artifact(opt)
if modeldir:
self.weights = Path(modeldir) / "last.pt"
config = self.wandb_run.config
opt.weights, opt.save_period, opt.batch_size, opt.bbox_interval, opt.epochs, opt.hyp = str(
self.weights), config.save_period, config.total_batch_size, config.bbox_interval, config.epochs, \
config.opt['hyp']
data_dict = dict(self.wandb_run.config.data_dict) # eliminates the need for config file to resume
if 'val_artifact' not in self.__dict__: # If --upload_dataset is set, use the existing artifact, don't download
self.train_artifact_path, self.train_artifact = self.download_dataset_artifact(data_dict.get('train'),
opt.artifact_alias)
self.val_artifact_path, self.val_artifact = self.download_dataset_artifact(data_dict.get('val'),
opt.artifact_alias)
self.result_artifact, self.result_table, self.val_table, self.weights = None, None, None, None
if self.train_artifact_path is not None:
train_path = Path(self.train_artifact_path) / 'data/images/'
data_dict['train'] = str(train_path)
if self.val_artifact_path is not None:
val_path = Path(self.val_artifact_path) / 'data/images/'
data_dict['val'] = str(val_path)
self.val_table = self.val_artifact.get("val")
self.map_val_table_path()
if self.val_artifact is not None:
self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
self.result_table = wandb.Table(["epoch", "id", "prediction", "avg_confidence"])
if opt.bbox_interval == -1:
self.bbox_interval = opt.bbox_interval = (opt.epochs // 10) if opt.epochs > 10 else 1
return data_dict
def download_dataset_artifact(self, path, alias):
if isinstance(path, str) and path.startswith(WANDB_ARTIFACT_PREFIX):
artifact_path = Path(remove_prefix(path, WANDB_ARTIFACT_PREFIX) + ":" + alias)
dataset_artifact = wandb.use_artifact(artifact_path.as_posix())
assert dataset_artifact is not None, "'Error: W&B dataset artifact doesn\'t exist'"
datadir = dataset_artifact.download()
return datadir, dataset_artifact
return None, None
def download_model_artifact(self, opt):
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
model_artifact = wandb.use_artifact(remove_prefix(opt.resume, WANDB_ARTIFACT_PREFIX) + ":latest")
assert model_artifact is not None, 'Error: W&B model artifact doesn\'t exist'
modeldir = model_artifact.download()
epochs_trained = model_artifact.metadata.get('epochs_trained')
total_epochs = model_artifact.metadata.get('total_epochs')
is_finished = total_epochs is None
assert not is_finished, 'training is finished, can only resume incomplete runs.'
return modeldir, model_artifact
return None, None
def log_model(self, path, opt, epoch, fitness_score, best_model=False):
model_artifact = wandb.Artifact('run_' + wandb.run.id + '_model', type='model', metadata={
'original_url': str(path),
'epochs_trained': epoch + 1,
'save period': opt.save_period,
'project': opt.project,
'total_epochs': opt.epochs,
'fitness_score': fitness_score
})
model_artifact.add_file(str(path / 'last.pt'), name='last.pt')
wandb.log_artifact(model_artifact,
aliases=['latest', 'last', 'epoch ' + str(self.current_epoch), 'best' if best_model else ''])
print("Saving model artifact on epoch ", epoch + 1)
def log_dataset_artifact(self, data_file, single_cls, project, overwrite_config=False):
with open(data_file) as f:
data = yaml.safe_load(f) # data dict
nc, names = (1, ['item']) if single_cls else (int(data['nc']), data['names'])
names = {k: v for k, v in enumerate(names)} # to index dictionary
self.train_artifact = self.create_dataset_table(LoadImagesAndLabels(
data['train'], rect=True, batch_size=1), names, name='train') if data.get('train') else None
self.val_artifact = self.create_dataset_table(LoadImagesAndLabels(
data['val'], rect=True, batch_size=1), names, name='val') if data.get('val') else None
if data.get('train'):
data['train'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'train')
if data.get('val'):
data['val'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'val')
path = data_file if overwrite_config else '_wandb.'.join(data_file.rsplit('.', 1)) # updated data.yaml path
data.pop('download', None)
with open(path, 'w') as f:
yaml.safe_dump(data, f)
if self.job_type == 'Training': # builds correct artifact pipeline graph
self.wandb_run.use_artifact(self.val_artifact)
self.wandb_run.use_artifact(self.train_artifact)
self.val_artifact.wait()
self.val_table = self.val_artifact.get('val')
self.map_val_table_path()
else:
self.wandb_run.log_artifact(self.train_artifact)
self.wandb_run.log_artifact(self.val_artifact)
return path
def map_val_table_path(self):
self.val_table_map = {}
print("Mapping dataset")
for i, data in enumerate(tqdm(self.val_table.data)):
self.val_table_map[data[3]] = data[0]
def create_dataset_table(self, dataset, class_to_id, name='dataset'):
# TODO: Explore multiprocessing to slpit this loop parallely| This is essential for speeding up the the logging
artifact = wandb.Artifact(name=name, type="dataset")
img_files = tqdm([dataset.path]) if isinstance(dataset.path, str) and Path(dataset.path).is_dir() else None
img_files = tqdm(dataset.img_files) if not img_files else img_files
for img_file in img_files:
if Path(img_file).is_dir():
artifact.add_dir(img_file, name='data/images')
labels_path = 'labels'.join(dataset.path.rsplit('images', 1))
artifact.add_dir(labels_path, name='data/labels')
else:
artifact.add_file(img_file, name='data/images/' + Path(img_file).name)
label_file = Path(img2label_paths([img_file])[0])
artifact.add_file(str(label_file),
name='data/labels/' + label_file.name) if label_file.exists() else None
table = wandb.Table(columns=["id", "train_image", "Classes", "name"])
class_set = wandb.Classes([{'id': id, 'name': name} for id, name in class_to_id.items()])
for si, (img, labels, paths, shapes) in enumerate(tqdm(dataset)):
box_data, img_classes = [], {}
for cls, *xywh in labels[:, 1:].tolist():
cls = int(cls)
box_data.append({"position": {"middle": [xywh[0], xywh[1]], "width": xywh[2], "height": xywh[3]},
"class_id": cls,
"box_caption": "%s" % (class_to_id[cls])})
img_classes[cls] = class_to_id[cls]
boxes = {"ground_truth": {"box_data": box_data, "class_labels": class_to_id}} # inference-space
table.add_data(si, wandb.Image(paths, classes=class_set, boxes=boxes), json.dumps(img_classes),
Path(paths).name)
artifact.add(table, name)
return artifact
def log_training_progress(self, predn, path, names):
if self.val_table and self.result_table:
class_set = wandb.Classes([{'id': id, 'name': name} for id, name in names.items()])
box_data = []
total_conf = 0
for *xyxy, conf, cls in predn.tolist():
if conf >= 0.25:
box_data.append(
{"position": {"minX": xyxy[0], "minY": xyxy[1], "maxX": xyxy[2], "maxY": xyxy[3]},
"class_id": int(cls),
"box_caption": "%s %.3f" % (names[cls], conf),
"scores": {"class_score": conf},
"domain": "pixel"})
total_conf = total_conf + conf
boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
id = self.val_table_map[Path(path).name]
self.result_table.add_data(self.current_epoch,
id,
wandb.Image(self.val_table.data[id][1], boxes=boxes, classes=class_set),
total_conf / max(1, len(box_data))
)
def log(self, log_dict):
if self.wandb_run:
for key, value in log_dict.items():
self.log_dict[key] = value
def end_epoch(self, best_result=False):
if self.wandb_run:
wandb.log(self.log_dict)
self.log_dict = {}
if self.result_artifact:
train_results = wandb.JoinedTable(self.val_table, self.result_table, "id")
self.result_artifact.add(train_results, 'result')
wandb.log_artifact(self.result_artifact, aliases=['latest', 'last', 'epoch ' + str(self.current_epoch),
('best' if best_result else '')])
self.result_table = wandb.Table(["epoch", "id", "prediction", "avg_confidence"])
self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
def finish_run(self):
if self.wandb_run:
if self.log_dict:
wandb.log(self.log_dict)
wandb.run.finish()