apoorva/greenhouse
2
1
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

yolo with runs.zip file exp 14 is the best weight

Browse Source
This commit is contained in:
Apoorva Gupta
2023-02-21 21:43:51 +05:30
parent 34abb2b0dd
commit dbe80aca78
43 changed files with 3670 additions and 5669 deletions
Download Patch File Download Diff File Expand all files Collapse all files
yolov3/utils/__init__.py
+6 -68
View File
@@ -3,78 +3,16 @@
utils/initialization
"""
import contextlib
import platform
import threading
def emojis(str=''):
# Return platform-dependent emoji-safe version of string
return str.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else str
class TryExcept(contextlib.ContextDecorator):
# YOLOv3 TryExcept class. Usage: @TryExcept() decorator or 'with TryExcept():' context manager
def __init__(self, msg=''):
self.msg = msg
def __enter__(self):
pass
def __exit__(self, exc_type, value, traceback):
if value:
print(emojis(f"{self.msg}{': ' if self.msg else ''}{value}"))
return True
def threaded(func):
# Multi-threads a target function and returns thread. Usage: @threaded decorator
def wrapper(*args, **kwargs):
thread = threading.Thread(target=func, args=args, kwargs=kwargs, daemon=True)
thread.start()
return thread
return wrapper
def join_threads(verbose=False):
# Join all daemon threads, i.e. atexit.register(lambda: join_threads())
main_thread = threading.current_thread()
for t in threading.enumerate():
if t is not main_thread:
if verbose:
print(f'Joining thread {t.name}')
t.join()
def notebook_init(verbose=True):
# Check system software and hardware
def notebook_init():
# For notebooks
print('Checking setup...')
import os
import shutil
from utils.general import check_font, check_requirements, is_colab
from utils.torch_utils import select_device # imports
check_font()
import psutil
from IPython import display # to display images and clear console output
if is_colab():
shutil.rmtree('/content/sample_data', ignore_errors=True) # remove colab /sample_data directory
# System info
if verbose:
gb = 1 << 30 # bytes to GiB (1024 ** 3)
ram = psutil.virtual_memory().total
total, used, free = shutil.disk_usage('/')
display.clear_output()
s = f'({os.cpu_count()} CPUs, {ram / gb:.1f} GB RAM, {(total - free) / gb:.1f}/{total / gb:.1f} GB disk)'
else:
s = ''
from utils.general import emojis
from utils.torch_utils import select_device # imports
display.clear_output()
select_device(newline=False)
print(emojis(f'Setup complete ✅ {s}'))
print(emojis('Setup complete ✅'))
return display
yolov3/utils/activations.py
+10 -12
View File
@@ -8,32 +8,29 @@ import torch.nn as nn
import torch.nn.functional as F
class SiLU(nn.Module):
# SiLU activation https://arxiv.org/pdf/1606.08415.pdf
# SiLU https://arxiv.org/pdf/1606.08415.pdf ----------------------------------------------------------------------------
class SiLU(nn.Module): # export-friendly version of nn.SiLU()
@staticmethod
def forward(x):
return x * torch.sigmoid(x)
class Hardswish(nn.Module):
# Hard-SiLU activation
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, 0.0, 6.0) / 6.0 # for TorchScript, CoreML and ONNX
# return x * F.hardsigmoid(x) # for torchscript and CoreML
return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0 # for torchscript, CoreML and ONNX
# Mish https://github.com/digantamisra98/Mish --------------------------------------------------------------------------
class Mish(nn.Module):
# Mish activation https://github.com/digantamisra98/Mish
@staticmethod
def forward(x):
return x * F.softplus(x).tanh()
class MemoryEfficientMish(nn.Module):
# Mish activation memory-efficient
class F(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
@@ -50,8 +47,8 @@ class MemoryEfficientMish(nn.Module):
return self.F.apply(x)
# FReLU https://arxiv.org/abs/2007.11824 -------------------------------------------------------------------------------
class FReLU(nn.Module):
# FReLU activation https://arxiv.org/abs/2007.11824
def __init__(self, c1, k=3): # ch_in, kernel
super().__init__()
self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False)
@@ -61,8 +58,9 @@ class FReLU(nn.Module):
return torch.max(x, self.bn(self.conv(x)))
# ACON https://arxiv.org/pdf/2009.04759.pdf ----------------------------------------------------------------------------
class AconC(nn.Module):
r""" ACON activation (activate or not)
r""" ACON activation (activate or not).
AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter
according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
"""
@@ -79,7 +77,7 @@ class AconC(nn.Module):
class MetaAconC(nn.Module):
r""" ACON activation (activate or not)
r""" ACON activation (activate or not).
MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network
according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
"""
yolov3/utils/augmentations.py
+17 -137
View File
@@ -8,42 +8,34 @@ import random
import cv2
import numpy as np
import torch
import torchvision.transforms as T
import torchvision.transforms.functional as TF
from utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box, xywhn2xyxy
from utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box
from utils.metrics import bbox_ioa
IMAGENET_MEAN = 0.485, 0.456, 0.406 # RGB mean
IMAGENET_STD = 0.229, 0.224, 0.225 # RGB standard deviation
class Albumentations:
# YOLOv3 Albumentations class (optional, only used if package is installed)
def __init__(self, size=640):
# Albumentations class (optional, only used if package is installed)
def __init__(self):
self.transform = None
prefix = colorstr('albumentations: ')
try:
import albumentations as A
check_version(A.__version__, '1.0.3', hard=True) # version requirement
T = [
A.RandomResizedCrop(height=size, width=size, scale=(0.8, 1.0), ratio=(0.9, 1.11), p=0.0),
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)] # transforms
self.transform = A.Compose(T, bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
A.ImageCompression(quality_lower=75, p=0.0)],
bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
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(f'{prefix}{e}')
LOGGER.info(colorstr('albumentations: ') + f'{e}')
def __call__(self, im, labels, p=1.0):
if self.transform and random.random() < p:
@@ -52,18 +44,6 @@ class Albumentations:
return im, labels
def normalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD, inplace=False):
# Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = (x - mean) / std
return TF.normalize(x, mean, std, inplace=inplace)
def denormalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD):
# Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = x * std + mean
for i in range(3):
x[:, i] = x[:, i] * std[i] + mean[i]
return x
def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
# HSV color-space augmentation
if hgain or sgain or vgain:
@@ -141,14 +121,7 @@ def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleF
return im, ratio, (dw, dh)
def random_perspective(im,
targets=(),
segments=(),
degrees=10,
translate=.1,
scale=.1,
shear=10,
perspective=0.0,
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]
@@ -201,7 +174,7 @@ def random_perspective(im,
# Transform label coordinates
n = len(targets)
if n:
use_segments = any(x.any() for x in segments) and len(segments) == 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
@@ -250,10 +223,12 @@ def copy_paste(im, labels, segments, p=0.5):
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, (1, 1, 1), cv2.FILLED)
cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED)
result = cv2.flip(im, 1) # augment segments (flip left-right)
i = cv2.flip(im_new, 1).astype(bool)
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
@@ -280,7 +255,7 @@ def cutout(im, labels, p=0.5):
# return unobscured labels
if len(labels) and s > 0.03:
box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
ioa = bbox_ioa(box, xywhn2xyxy(labels[:, 1:5], w, h)) # intersection over area
ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
labels = labels[ioa < 0.60] # remove >60% obscured labels
return labels
@@ -294,104 +269,9 @@ def mixup(im, labels, im2, labels2):
return im, labels
def box_candidates(box1, box2, wh_thr=2, ar_thr=100, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n)
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
def classify_albumentations(
augment=True,
size=224,
scale=(0.08, 1.0),
ratio=(0.75, 1.0 / 0.75), # 0.75, 1.33
hflip=0.5,
vflip=0.0,
jitter=0.4,
mean=IMAGENET_MEAN,
std=IMAGENET_STD,
auto_aug=False):
# YOLOv3 classification Albumentations (optional, only used if package is installed)
prefix = colorstr('albumentations: ')
try:
import albumentations as A
from albumentations.pytorch import ToTensorV2
check_version(A.__version__, '1.0.3', hard=True) # version requirement
if augment: # Resize and crop
T = [A.RandomResizedCrop(height=size, width=size, scale=scale, ratio=ratio)]
if auto_aug:
# TODO: implement AugMix, AutoAug & RandAug in albumentation
LOGGER.info(f'{prefix}auto augmentations are currently not supported')
else:
if hflip > 0:
T += [A.HorizontalFlip(p=hflip)]
if vflip > 0:
T += [A.VerticalFlip(p=vflip)]
if jitter > 0:
color_jitter = (float(jitter),) * 3 # repeat value for brightness, contrast, satuaration, 0 hue
T += [A.ColorJitter(*color_jitter, 0)]
else: # Use fixed crop for eval set (reproducibility)
T = [A.SmallestMaxSize(max_size=size), A.CenterCrop(height=size, width=size)]
T += [A.Normalize(mean=mean, std=std), ToTensorV2()] # Normalize and convert to Tensor
LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
return A.Compose(T)
except ImportError: # package not installed, skip
LOGGER.warning(f'{prefix}⚠️ not found, install with `pip install albumentations` (recommended)')
except Exception as e:
LOGGER.info(f'{prefix}{e}')
def classify_transforms(size=224):
# Transforms to apply if albumentations not installed
assert isinstance(size, int), f'ERROR: classify_transforms size {size} must be integer, not (list, tuple)'
# T.Compose([T.ToTensor(), T.Resize(size), T.CenterCrop(size), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
return T.Compose([CenterCrop(size), ToTensor(), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
class LetterBox:
# YOLOv3 LetterBox class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
def __init__(self, size=(640, 640), auto=False, stride=32):
super().__init__()
self.h, self.w = (size, size) if isinstance(size, int) else size
self.auto = auto # pass max size integer, automatically solve for short side using stride
self.stride = stride # used with auto
def __call__(self, im): # im = np.array HWC
imh, imw = im.shape[:2]
r = min(self.h / imh, self.w / imw) # ratio of new/old
h, w = round(imh * r), round(imw * r) # resized image
hs, ws = (math.ceil(x / self.stride) * self.stride for x in (h, w)) if self.auto else self.h, self.w
top, left = round((hs - h) / 2 - 0.1), round((ws - w) / 2 - 0.1)
im_out = np.full((self.h, self.w, 3), 114, dtype=im.dtype)
im_out[top:top + h, left:left + w] = cv2.resize(im, (w, h), interpolation=cv2.INTER_LINEAR)
return im_out
class CenterCrop:
# YOLOv3 CenterCrop class for image preprocessing, i.e. T.Compose([CenterCrop(size), ToTensor()])
def __init__(self, size=640):
super().__init__()
self.h, self.w = (size, size) if isinstance(size, int) else size
def __call__(self, im): # im = np.array HWC
imh, imw = im.shape[:2]
m = min(imh, imw) # min dimension
top, left = (imh - m) // 2, (imw - m) // 2
return cv2.resize(im[top:top + m, left:left + m], (self.w, self.h), interpolation=cv2.INTER_LINEAR)
class ToTensor:
# YOLOv3 ToTensor class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
def __init__(self, half=False):
super().__init__()
self.half = half
def __call__(self, im): # im = np.array HWC in BGR order
im = np.ascontiguousarray(im.transpose((2, 0, 1))[::-1]) # HWC to CHW -> BGR to RGB -> contiguous
im = torch.from_numpy(im) # to torch
im = im.half() if self.half else im.float() # uint8 to fp16/32
im /= 255.0 # 0-255 to 0.0-1.0
return im
yolov3/utils/autoanchor.py
+32 -37
View File
@@ -1,6 +1,6 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
"""
AutoAnchor utils
Auto-anchor utils
"""
import random
@@ -10,23 +10,21 @@ import torch
import yaml
from tqdm import tqdm
from utils import TryExcept
from utils.general import LOGGER, TQDM_BAR_FORMAT, 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.anchors.prod(-1).mean(-1).view(-1) # mean anchor area per output layer
# 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 and (da.sign() != ds.sign()): # same order
if da.sign() != ds.sign(): # same order
LOGGER.info(f'{PREFIX}Reversing anchor order')
m.anchors[:] = m.anchors.flip(0)
@TryExcept(f'{PREFIX}ERROR')
def check_anchors(dataset, model, thr=4.0, imgsz=640):
# Check anchor fit to data, recompute if necessary
m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect()
@@ -42,26 +40,26 @@ def check_anchors(dataset, model, thr=4.0, imgsz=640):
bpr = (best > 1 / thr).float().mean() # best possible recall
return bpr, aat
stride = m.stride.to(m.anchors.device).view(-1, 1, 1) # model strides
anchors = m.anchors.clone() * stride # current 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(f'{s}Current anchors are a good fit to dataset ✅')
LOGGER.info(emojis(f'{s}Current anchors are a good fit to dataset ✅'))
else:
LOGGER.info(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...')
LOGGER.info(emojis(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...'))
na = m.anchors.numel() // 2 # number of anchors
anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
try:
anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
except Exception as 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.anchors[:] = anchors.clone().view_as(m.anchors)
check_anchor_order(m) # must be in pixel-space (not grid-space)
m.anchors /= stride
s = f'{PREFIX}Done ✅ (optional: update model *.yaml to use these anchors in the future)'
m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss
check_anchor_order(m)
LOGGER.info(f'{PREFIX}New anchors saved to model. Update model *.yaml to use these anchors in the future.')
else:
s = f'{PREFIX}Done ⚠️ (original anchors better than new anchors, proceeding with original anchors)'
LOGGER.info(s)
LOGGER.info(f'{PREFIX}Original anchors better than new anchors. Proceeding with original anchors.')
def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
@@ -83,7 +81,6 @@ def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen
"""
from scipy.cluster.vq import kmeans
npr = np.random
thr = 1 / thr
def metric(k, wh): # compute metrics
@@ -103,7 +100,7 @@ def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen
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 x in k:
for i, x in enumerate(k):
s += '%i,%i, ' % (round(x[0]), round(x[1]))
if verbose:
LOGGER.info(s[:-2])
@@ -112,7 +109,7 @@ def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen
if isinstance(dataset, str): # *.yaml file
with open(dataset, errors='ignore') as f:
data_dict = yaml.safe_load(f) # model dict
from utils.dataloaders import LoadImagesAndLabels
from utils.datasets import LoadImagesAndLabels
dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
# Get label wh
@@ -122,21 +119,18 @@ def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen
# Filter
i = (wh0 < 3.0).any(1).sum()
if i:
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)].astype(np.float32) # filter > 2 pixels
# wh = wh * (npr.rand(wh.shape[0], 1) * 0.9 + 0.1) # multiply by random scale 0-1
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 init
try:
LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(wh)} points...')
assert n <= len(wh) # apply overdetermined constraint
s = wh.std(0) # sigmas for whitening
k = kmeans(wh / s, n, iter=30)[0] * s # points
assert n == len(k) # kmeans may return fewer points than requested if wh is insufficient or too similar
except Exception:
LOGGER.warning(f'{PREFIX}WARNING ⚠️ switching strategies from kmeans to random init')
k = np.sort(npr.rand(n * 2)).reshape(n, 2) * img_size # random init
wh, wh0 = (torch.tensor(x, dtype=torch.float32) for x in (wh, wh0))
# Kmeans calculation
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, 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, verbose=False)
# Plot
@@ -152,8 +146,9 @@ def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen
# fig.savefig('wh.png', dpi=200)
# 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), bar_format=TQDM_BAR_FORMAT) # 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)
@@ -166,4 +161,4 @@ def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen
if verbose:
print_results(k, verbose)
return print_results(k).astype(np.float32)
return print_results(k)
yolov3/utils/autobatch.py
+17 -32
View File
@@ -7,66 +7,51 @@ 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, amp=True):
# Check YOLOv3 training batch size
with torch.cuda.amp.autocast(amp):
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=0.8, batch_size=16):
# Automatically estimate best YOLOv3 batch size to use `fraction` of available CUDA memory
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/yolov5', 'yolov5s', autoshape=False)
# model = torch.hub.load('ultralytics/yolov3', 'yolov3', autoshape=False)
# print(autobatch(model))
# Check device
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
if torch.backends.cudnn.benchmark:
LOGGER.info(f'{prefix} ⚠️ Requires torch.backends.cudnn.benchmark=False, using default batch-size {batch_size}')
return batch_size
# Inspect CUDA memory
gb = 1 << 30 # bytes to GiB (1024 ** 3)
d = str(device).upper() # 'CUDA:0'
properties = torch.cuda.get_device_properties(device) # device properties
t = properties.total_memory / gb # GiB total
r = torch.cuda.memory_reserved(device) / gb # GiB reserved
a = torch.cuda.memory_allocated(device) / gb # GiB allocated
f = t - (r + a) # GiB free
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')
# Profile batch sizes
batch_sizes = [1, 2, 4, 8, 16]
try:
img = [torch.empty(b, 3, imgsz, imgsz) for b in batch_sizes]
results = profile(img, model, n=3, device=device)
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}')
# Fit a solution
y = [x[2] for x in results if x] # memory [2]
p = np.polyfit(batch_sizes[:len(y)], y, deg=1) # first degree polynomial fit
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)
if None in results: # some sizes failed
i = results.index(None) # first fail index
if b >= batch_sizes[i]: # y intercept above failure point
b = batch_sizes[max(i - 1, 0)] # select prior safe point
if b < 1 or b > 1024: # b outside of safe range
b = batch_size
LOGGER.warning(f'{prefix}WARNING ⚠️ CUDA anomaly detected, recommend restart environment and retry command.')
fraction = (np.polyval(p, b) + r + a) / t # actual fraction predicted
LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%) ✅')
LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%)')
return b
yolov3/utils/callbacks.py
+41 -41
View File
@@ -3,46 +3,46 @@
Callback utils
"""
import threading
class Callbacks:
""""
Handles all registered callbacks for YOLOv3 Hooks
Handles all registered callbacks for Hooks
"""
def __init__(self):
# Define the available callbacks
self._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': [],
'on_params_update': [],
'teardown': [],}
self.stop_training = False # set True to interrupt training
# 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
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"
@@ -53,24 +53,24 @@ class Callbacks:
Returns all the registered actions by callback hook
Args:
hook: The name of the hook to check, defaults to all
hook The name of the hook to check, defaults to all
"""
return self._callbacks[hook] if hook else self._callbacks
if hook:
return self._callbacks[hook]
else:
return self._callbacks
def run(self, hook, *args, thread=False, **kwargs):
def run(self, hook, *args, **kwargs):
"""
Loop through the registered actions and fire all callbacks on main thread
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 YOLOv3
thread: (boolean) Run callbacks in daemon thread
kwargs: Keyword Arguments to receive from YOLOv3
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]:
if thread:
threading.Thread(target=logger['callback'], args=args, kwargs=kwargs, daemon=True).start()
else:
logger['callback'](*args, **kwargs)
logger['callback'](*args, **kwargs)
yolov3/utils/downloads.py
+104 -61
View File
@@ -3,104 +3,147 @@
Download utils
"""
import logging
import os
import platform
import subprocess
import time
import urllib
from pathlib import Path
from zipfile import ZipFile
import requests
import torch
def is_url(url, check=True):
# Check if string is URL and check if URL exists
try:
url = str(url)
result = urllib.parse.urlparse(url)
assert all([result.scheme, result.netloc]) # check if is url
return (urllib.request.urlopen(url).getcode() == 200) if check else True # check if exists online
except (AssertionError, urllib.request.HTTPError):
return False
def gsutil_getsize(url=''):
# gs://bucket/file size https://cloud.google.com/storage/docs/gsutil/commands/du
s = subprocess.check_output(f'gsutil du {url}', shell=True).decode('utf-8')
return eval(s.split(' ')[0]) if len(s) else 0 # bytes
def url_getsize(url='https://ultralytics.com/images/bus.jpg'):
# Return downloadable file size in bytes
response = requests.head(url, allow_redirects=True)
return int(response.headers.get('content-length', -1))
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
from utils.general import LOGGER
file = Path(file)
assert_msg = f"Downloaded file '{file}' does not exist or size is < min_bytes={min_bytes}"
try: # url1
LOGGER.info(f'Downloading {url} to {file}...')
torch.hub.download_url_to_file(url, str(file), progress=LOGGER.level <= logging.INFO)
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, assert_msg # check
except Exception as e: # url2
if file.exists():
file.unlink() # remove partial downloads
LOGGER.info(f'ERROR: {e}\nRe-attempting {url2 or url} to {file}...')
subprocess.run(
f"curl -# -L '{url2 or url}' -o '{file}' --retry 3 -C -".split()) # curl download, retry and resume on fail
file.unlink(missing_ok=True) # remove partial downloads
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
if file.exists():
file.unlink() # remove partial downloads
LOGGER.info(f'ERROR: {assert_msg}\n{error_msg}')
LOGGER.info('')
file.unlink(missing_ok=True) # remove partial downloads
print(f"ERROR: {assert_msg}\n{error_msg}")
print('')
def attempt_download(file, repo='ultralytics/yolov5', release='v7.0'):
# Attempt file download from GitHub release assets if not found locally. release = 'latest', 'v7.0', etc.
from utils.general import LOGGER
def github_assets(repository, version='latest'):
# Return GitHub repo tag (i.e. 'v7.0') and assets (i.e. ['yolov5s.pt', 'yolov5m.pt', ...])
if version != 'latest':
version = f'tags/{version}' # i.e. tags/v7.0
response = requests.get(f'https://api.github.com/repos/{repository}/releases/{version}').json() # github api
return response['tag_name'], [x['name'] for x in response['assets']] # tag, assets
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 = Path(urllib.parse.unquote(str(file))).name # decode '%2F' to '/' etc.
if str(file).startswith(('http:/', 'https:/')): # download
url = str(file).replace(':/', '://') # Pathlib turns :// -> :/
file = name.split('?')[0] # parse authentication https://url.com/file.txt?auth...
if Path(file).is_file():
LOGGER.info(f'Found {url} locally at {file}') # file already exists
else:
safe_download(file=file, url=url, min_bytes=1E5)
return file
name = name.split('?')[0] # parse authentication https://url.com/file.txt?auth...
safe_download(file=name, url=url, min_bytes=1E5)
return name
# GitHub assets
assets = [f'yolov5{size}{suffix}.pt' for size in 'nsmlx' for suffix in ('', '6', '-cls', '-seg')] # default
try:
tag, assets = github_assets(repo, release)
except Exception:
try:
tag, assets = github_assets(repo) # latest release
except Exception:
try:
tag = subprocess.check_output('git tag', shell=True, stderr=subprocess.STDOUT).decode().split()[-1]
except Exception:
tag = release
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. ['yolov3.pt'...]
tag = response['tag_name'] # i.e. 'v1.0'
except: # fallback plan
assets = ['yolov3.pt', 'yolov3-spp.pt', 'yolov3-tiny.pt']
try:
tag = subprocess.check_output('git tag', shell=True, stderr=subprocess.STDOUT).decode().split()[-1]
except:
tag = 'v9.5.0' # current release
if name in assets:
safe_download(file,
url=f'https://github.com/{repo}/releases/download/{tag}/{name}',
# url2=f'https://storage.googleapis.com/{repo}/ckpt/{name}', # backup url (optional)
min_bytes=1E5,
error_msg=f'{file} missing, try downloading from https://github.com/{repo}/releases/{tag}')
error_msg=f'{file} missing, try downloading from https://github.com/{repo}/releases/')
return str(file)
def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'):
# Downloads a file from Google Drive. from yolov3.utils.downloads import *; gdrive_download()
t = time.time()
file = Path(file)
cookie = Path('cookie') # gdrive cookie
print(f'Downloading https://drive.google.com/uc?export=download&id={id} as {file}... ', end='')
file.unlink(missing_ok=True) # remove existing file
cookie.unlink(missing_ok=True) # remove existing cookie
# Attempt file download
out = "NUL" if platform.system() == "Windows" else "/dev/null"
os.system(f'curl -c ./cookie -s -L "drive.google.com/uc?export=download&id={id}" > {out}')
if os.path.exists('cookie'): # large file
s = f'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm={get_token()}&id={id}" -o {file}'
else: # small file
s = f'curl -s -L -o {file} "drive.google.com/uc?export=download&id={id}"'
r = os.system(s) # execute, capture return
cookie.unlink(missing_ok=True) # remove existing cookie
# Error check
if r != 0:
file.unlink(missing_ok=True) # remove partial
print('Download error ') # raise Exception('Download error')
return r
# Unzip if archive
if file.suffix == '.zip':
print('unzipping... ', end='')
ZipFile(file).extractall(path=file.parent) # unzip
file.unlink() # remove zip
print(f'Done ({time.time() - t:.1f}s)')
return r
def get_token(cookie="./cookie"):
with open(cookie) as f:
for line in f:
if "download" in line:
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
#
# storage_client = storage.Client()
# bucket = storage_client.get_bucket(bucket_name)
# blob = bucket.blob(destination_blob_name)
#
# blob.upload_from_filename(source_file_name)
#
# print('File {} uploaded to {}.'.format(
# source_file_name,
# destination_blob_name))
#
#
# def download_blob(bucket_name, source_blob_name, destination_file_name):
# # Uploads a blob from a bucket
# storage_client = storage.Client()
# bucket = storage_client.get_bucket(bucket_name)
# blob = bucket.blob(source_blob_name)
#
# blob.download_to_filename(destination_file_name)
#
# print('Blob {} downloaded to {}.'.format(
# source_blob_name,
# destination_file_name))
yolov3/utils/general.py
Regular → Executable
+337 -631
View File
File diff suppressed because it is too large Load Diff
yolov3/utils/loggers/__init__.py
+52 -294
View File
@@ -5,26 +5,25 @@ Logging utils
import os
import warnings
from pathlib import Path
from threading import Thread
import pkg_resources as pkg
import torch
from torch.utils.tensorboard import SummaryWriter
from utils.general import LOGGER, colorstr, cv2
from utils.loggers.clearml.clearml_utils import ClearmlLogger
from utils.general import colorstr, emojis
from utils.loggers.wandb.wandb_utils import WandbLogger
from utils.plots import plot_images, plot_labels, plot_results
from utils.plots import plot_images, plot_results
from utils.torch_utils import de_parallel
LOGGERS = ('csv', 'tb', 'wandb', 'clearml', 'comet') # *.csv, TensorBoard, Weights & Biases, ClearML
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}:
if pkg.parse_version(wandb.__version__) >= pkg.parse_version('0.12.2') and RANK in [0, -1]:
try:
wandb_login_success = wandb.login(timeout=30)
except wandb.errors.UsageError: # known non-TTY terminal issue
@@ -34,64 +33,30 @@ try:
except (ImportError, AssertionError):
wandb = None
try:
import clearml
assert hasattr(clearml, '__version__') # verify package import not local dir
except (ImportError, AssertionError):
clearml = None
try:
if RANK not in [0, -1]:
comet_ml = None
else:
import comet_ml
assert hasattr(comet_ml, '__version__') # verify package import not local dir
from utils.loggers.comet import CometLogger
except (ModuleNotFoundError, ImportError, AssertionError):
comet_ml = None
class Loggers():
# YOLOv3 Loggers class
# 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.plots = not opt.noplots # plot results
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
self.best_keys = ['best/epoch', 'best/precision', 'best/recall', 'best/mAP_0.5', 'best/mAP_0.5:0.95']
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
# Messages
if not clearml:
prefix = colorstr('ClearML: ')
s = f"{prefix}run 'pip install clearml' to automatically track, visualize and remotely train YOLOv3 🚀 in ClearML"
self.logger.info(s)
if not comet_ml:
prefix = colorstr('Comet: ')
s = f"{prefix}run 'pip install comet_ml' to automatically track and visualize YOLOv3 🚀 runs in Comet"
self.logger.info(s)
# 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:
@@ -101,127 +66,53 @@ class Loggers():
# 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)
self.wandb = WandbLogger(self.opt, run_id)
else:
self.wandb = None
# ClearML
if clearml and 'clearml' in self.include:
try:
self.clearml = ClearmlLogger(self.opt, self.hyp)
except Exception:
self.clearml = None
prefix = colorstr('ClearML: ')
LOGGER.warning(f'{prefix}WARNING ⚠️ ClearML is installed but not configured, skipping ClearML logging.'
f' See https://github.com/ultralytics/yolov5/tree/master/utils/loggers/clearml#readme')
else:
self.clearml = None
# Comet
if comet_ml and 'comet' in self.include:
if isinstance(self.opt.resume, str) and self.opt.resume.startswith('comet://'):
run_id = self.opt.resume.split('/')[-1]
self.comet_logger = CometLogger(self.opt, self.hyp, run_id=run_id)
else:
self.comet_logger = CometLogger(self.opt, self.hyp)
else:
self.comet_logger = None
@property
def remote_dataset(self):
# Get data_dict if custom dataset artifact link is provided
data_dict = None
if self.clearml:
data_dict = self.clearml.data_dict
if self.wandb:
data_dict = self.wandb.data_dict
if self.comet_logger:
data_dict = self.comet_logger.data_dict
return data_dict
def on_train_start(self):
if self.comet_logger:
self.comet_logger.on_train_start()
def on_pretrain_routine_start(self):
if self.comet_logger:
self.comet_logger.on_pretrain_routine_start()
def on_pretrain_routine_end(self, labels, names):
def on_pretrain_routine_end(self):
# Callback runs on pre-train routine end
if self.plots:
plot_labels(labels, names, self.save_dir)
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]})
# if self.clearml:
# pass # ClearML saves these images automatically using hooks
if self.comet_logger:
self.comet_logger.on_pretrain_routine_end(paths)
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, model, ni, imgs, targets, paths, vals):
log_dict = dict(zip(self.keys[:3], vals))
def on_train_batch_end(self, ni, model, imgs, targets, paths, plots, sync_bn):
# Callback runs on train batch end
# ni: number integrated batches (since train start)
if self.plots:
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
plot_images(imgs, targets, paths, f)
if ni == 0 and self.tb and not self.opt.sync_bn:
log_tensorboard_graph(self.tb, model, imgsz=(self.opt.imgsz, self.opt.imgsz))
if ni == 10 and (self.wandb or self.clearml):
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'))
if self.wandb:
self.wandb.log({'Mosaics': [wandb.Image(str(f), caption=f.name) for f in files if f.exists()]})
if self.clearml:
self.clearml.log_debug_samples(files, title='Mosaics')
if self.comet_logger:
self.comet_logger.on_train_batch_end(log_dict, step=ni)
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
if self.comet_logger:
self.comet_logger.on_train_epoch_end(epoch)
def on_val_start(self):
if self.comet_logger:
self.comet_logger.on_val_start()
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)
if self.clearml:
self.clearml.log_image_with_boxes(path, pred, names, im)
def on_val_batch_end(self, batch_i, im, targets, paths, shapes, out):
if self.comet_logger:
self.comet_logger.on_val_batch_end(batch_i, im, targets, paths, shapes, out)
def on_val_end(self, nt, tp, fp, p, r, f1, ap, ap50, ap_class, confusion_matrix):
def on_val_end(self):
# Callback runs on val end
if self.wandb or self.clearml:
files = sorted(self.save_dir.glob('val*.jpg'))
if self.wandb:
self.wandb.log({'Validation': [wandb.Image(str(f), caption=f.name) for f in files]})
if self.clearml:
self.clearml.log_debug_samples(files, title='Validation')
if self.comet_logger:
self.comet_logger.on_val_end(nt, tp, fp, p, r, f1, ap, ap50, ap_class, confusion_matrix)
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 = dict(zip(self.keys, vals))
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
@@ -232,170 +123,37 @@ class Loggers():
if self.tb:
for k, v in x.items():
self.tb.add_scalar(k, v, epoch)
elif self.clearml: # log to ClearML if TensorBoard not used
for k, v in x.items():
title, series = k.split('/')
self.clearml.task.get_logger().report_scalar(title, series, v, epoch)
if self.wandb:
if best_fitness == fi:
best_results = [epoch] + vals[3:7]
for i, name in enumerate(self.best_keys):
self.wandb.wandb_run.summary[name] = best_results[i] # log best results in the summary
self.wandb.log(x)
self.wandb.end_epoch()
if self.clearml:
self.clearml.current_epoch_logged_images = set() # reset epoch image limit
self.clearml.current_epoch += 1
if self.comet_logger:
self.comet_logger.on_fit_epoch_end(x, epoch=epoch)
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 (epoch + 1) % self.opt.save_period == 0 and not final_epoch and self.opt.save_period != -1:
if self.wandb:
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)
if self.clearml:
self.clearml.task.update_output_model(model_path=str(last),
model_name='Latest Model',
auto_delete_file=False)
if self.comet_logger:
self.comet_logger.on_model_save(last, epoch, final_epoch, best_fitness, fi)
def on_train_end(self, last, best, epoch, results):
# Callback runs on training end, i.e. saving best model
if self.plots:
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
self.logger.info(f"Results saved to {colorstr('bold', self.save_dir)}")
if self.tb and not self.clearml: # These images are already captured by ClearML by now, we don't want doubles
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(dict(zip(self.keys[3:10], results)))
self.wandb.log({'Results': [wandb.Image(str(f), caption=f.name) for f in files]})
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=f'run_{self.wandb.wandb_run.id}_model',
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()
if self.clearml and not self.opt.evolve:
self.clearml.task.update_output_model(model_path=str(best if best.exists() else last),
name='Best Model',
auto_delete_file=False)
if self.comet_logger:
final_results = dict(zip(self.keys[3:10], results))
self.comet_logger.on_train_end(files, self.save_dir, last, best, epoch, final_results)
def on_params_update(self, params: dict):
# Update hyperparams or configs of the experiment
if self.wandb:
self.wandb.wandb_run.config.update(params, allow_val_change=True)
if self.comet_logger:
self.comet_logger.on_params_update(params)
class GenericLogger:
"""
YOLOv5 General purpose logger for non-task specific logging
Usage: from utils.loggers import GenericLogger; logger = GenericLogger(...)
Arguments
opt: Run arguments
console_logger: Console logger
include: loggers to include
"""
def __init__(self, opt, console_logger, include=('tb', 'wandb')):
# init default loggers
self.save_dir = Path(opt.save_dir)
self.include = include
self.console_logger = console_logger
self.csv = self.save_dir / 'results.csv' # CSV logger
if 'tb' in self.include:
prefix = colorstr('TensorBoard: ')
self.console_logger.info(
f"{prefix}Start with 'tensorboard --logdir {self.save_dir.parent}', view at http://localhost:6006/")
self.tb = SummaryWriter(str(self.save_dir))
if wandb and 'wandb' in self.include:
self.wandb = wandb.init(project=web_project_name(str(opt.project)),
name=None if opt.name == 'exp' else opt.name,
config=opt)
else:
self.wandb = None
def log_metrics(self, metrics, epoch):
# Log metrics dictionary to all loggers
if self.csv:
keys, vals = list(metrics.keys()), list(metrics.values())
n = len(metrics) + 1 # number of cols
s = '' if self.csv.exists() else (('%23s,' * n % tuple(['epoch'] + keys)).rstrip(',') + '\n') # header
with open(self.csv, 'a') as f:
f.write(s + ('%23.5g,' * n % tuple([epoch] + vals)).rstrip(',') + '\n')
if self.tb:
for k, v in metrics.items():
self.tb.add_scalar(k, v, epoch)
if self.wandb:
self.wandb.log(metrics, step=epoch)
def log_images(self, files, name='Images', epoch=0):
# Log images to all loggers
files = [Path(f) for f in (files if isinstance(files, (tuple, list)) else [files])] # to Path
files = [f for f in files if f.exists()] # filter by exists
if self.tb:
for f in files:
self.tb.add_image(f.stem, cv2.imread(str(f))[..., ::-1], epoch, dataformats='HWC')
if self.wandb:
self.wandb.log({name: [wandb.Image(str(f), caption=f.name) for f in files]}, step=epoch)
def log_graph(self, model, imgsz=(640, 640)):
# Log model graph to all loggers
if self.tb:
log_tensorboard_graph(self.tb, model, imgsz)
def log_model(self, model_path, epoch=0, metadata={}):
# Log model to all loggers
if self.wandb:
art = wandb.Artifact(name=f'run_{wandb.run.id}_model', type='model', metadata=metadata)
art.add_file(str(model_path))
wandb.log_artifact(art)
def update_params(self, params):
# Update the parameters logged
if self.wandb:
wandb.run.config.update(params, allow_val_change=True)
def log_tensorboard_graph(tb, model, imgsz=(640, 640)):
# Log model graph to TensorBoard
try:
p = next(model.parameters()) # for device, type
imgsz = (imgsz, imgsz) if isinstance(imgsz, int) else imgsz # expand
im = torch.zeros((1, 3, *imgsz)).to(p.device).type_as(p) # input image (WARNING: must be zeros, not empty)
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress jit trace warning
tb.add_graph(torch.jit.trace(de_parallel(model), im, strict=False), [])
except Exception as e:
LOGGER.warning(f'WARNING ⚠️ TensorBoard graph visualization failure {e}')
def web_project_name(project):
# Convert local project name to web project name
if not project.startswith('runs/train'):
return project
suffix = '-Classify' if project.endswith('-cls') else '-Segment' if project.endswith('-seg') else ''
return f'YOLOv5{suffix}'
self.wandb.finish_run()
else:
self.wandb.finish_run()
self.wandb = WandbLogger(self.opt)
yolov3/utils/loggers/wandb/wandb_utils.py
+382 -43
View File
@@ -1,32 +1,108 @@
# YOLOv3 🚀 by Ultralytics, GPL-3.0 license
# WARNING ⚠️ wandb is deprecated and will be removed in future release.
# See supported integrations at https://github.com/ultralytics/yolov5#integrations
"""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
from utils.general import LOGGER, colorstr
import pkg_resources as pkg
import yaml
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[3] # YOLOv5 root directory
ROOT = FILE.parents[3] # root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
RANK = int(os.getenv('RANK', -1))
DEPRECATION_WARNING = f"{colorstr('wandb')}: WARNING ⚠️ wandb is deprecated and will be removed in a future release. " \
f'See supported integrations at https://github.com/ultralytics/yolov5#integrations.'
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
LOGGER.warning(DEPRECATION_WARNING)
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.
@@ -46,7 +122,7 @@ class WandbLogger():
"""
- Initialize WandbLogger instance
- Upload dataset if opt.upload_dataset is True
- Setup training processes if job_type is 'Training'
- Setup trainig processes if job_type is 'Training'
arguments:
opt (namespace) -- Commandline arguments for this run
@@ -56,31 +132,82 @@ class WandbLogger():
"""
# Pre-training routine --
self.job_type = job_type
self.wandb, self.wandb_run = wandb, wandb.run if wandb else None
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
if self.wandb:
# 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='YOLOv5' if opt.project == 'runs/train' else Path(opt.project).stem,
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 isinstance(opt.data, dict):
# This means another dataset manager has already processed the dataset info (e.g. ClearML)
# and they will have stored the already processed dict in opt.data
self.data_dict = opt.data
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:
@@ -95,18 +222,77 @@ class WandbLogger():
self.log_dict, self.current_epoch = {}, 0
self.bbox_interval = opt.bbox_interval
if isinstance(opt.resume, str):
model_dir, _ = self.download_model_artifact(opt)
if model_dir:
self.weights = Path(model_dir) / 'last.pt'
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, opt.imgsz = str(
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, config.imgsz
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
if opt.evolve or opt.noplots:
self.bbox_interval = opt.bbox_interval = opt.epochs + 1 # disable bbox_interval
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):
"""
@@ -119,22 +305,166 @@ class WandbLogger():
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 = 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}')
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):
pass
"""
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):
"""
@@ -147,7 +477,7 @@ class WandbLogger():
for key, value in log_dict.items():
self.log_dict[key] = value
def end_epoch(self):
def end_epoch(self, best_result=False):
"""
commit the log_dict, model artifacts and Tables to W&B and flush the log_dict.
@@ -156,15 +486,25 @@ class WandbLogger():
"""
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}'
)
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):
"""
@@ -175,7 +515,6 @@ class WandbLogger():
with all_logging_disabled():
wandb.log(self.log_dict)
wandb.run.finish()
LOGGER.warning(DEPRECATION_WARNING)
@contextmanager
yolov3/utils/metrics.py
+86 -115
View File
@@ -11,8 +11,6 @@ import matplotlib.pyplot as plt
import numpy as np
import torch
from utils import TryExcept, threaded
def fitness(x):
# Model fitness as a weighted combination of metrics
@@ -20,15 +18,7 @@ def fitness(x):
return (x[:, :4] * w).sum(1)
def smooth(y, f=0.05):
# Box filter of fraction f
nf = round(len(y) * f * 2) // 2 + 1 # number of filter elements (must be odd)
p = np.ones(nf // 2) # ones padding
yp = np.concatenate((p * y[0], y, p * y[-1]), 0) # y padded
return np.convolve(yp, np.ones(nf) / nf, mode='valid') # y-smoothed
def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=(), eps=1e-16, prefix=''):
def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=()):
""" Compute the average precision, given the recall and precision curves.
Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
# Arguments
@@ -47,7 +37,7 @@ def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names
tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]
# Find unique classes
unique_classes, nt = np.unique(target_cls, return_counts=True)
unique_classes = np.unique(target_cls)
nc = unique_classes.shape[0] # number of classes, number of detections
# Create Precision-Recall curve and compute AP for each class
@@ -55,44 +45,42 @@ def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names
ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000))
for ci, c in enumerate(unique_classes):
i = pred_cls == c
n_l = nt[ci] # number of labels
n_l = (target_cls == c).sum() # number of labels
n_p = i.sum() # number of predictions
if n_p == 0 or n_l == 0:
continue
else:
# Accumulate FPs and TPs
fpc = (1 - tp[i]).cumsum(0)
tpc = tp[i].cumsum(0)
# Accumulate FPs and TPs
fpc = (1 - tp[i]).cumsum(0)
tpc = tp[i].cumsum(0)
# Recall
recall = tpc / (n_l + 1e-16) # recall curve
r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases
# Recall
recall = tpc / (n_l + eps) # recall curve
r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases
# Precision
precision = tpc / (tpc + fpc) # precision curve
p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score
# Precision
precision = tpc / (tpc + fpc) # precision curve
p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score
# AP from recall-precision curve
for j in range(tp.shape[1]):
ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
if plot and j == 0:
py.append(np.interp(px, mrec, mpre)) # precision at mAP@0.5
# AP from recall-precision curve
for j in range(tp.shape[1]):
ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
if plot and j == 0:
py.append(np.interp(px, mrec, mpre)) # precision at mAP@0.5
# Compute F1 (harmonic mean of precision and recall)
f1 = 2 * p * r / (p + r + eps)
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 = dict(enumerate(names)) # to dict
names = {i: v for i, v in enumerate(names)} # to dict
if plot:
plot_pr_curve(px, py, ap, Path(save_dir) / f'{prefix}PR_curve.png', names)
plot_mc_curve(px, f1, Path(save_dir) / f'{prefix}F1_curve.png', names, ylabel='F1')
plot_mc_curve(px, p, Path(save_dir) / f'{prefix}P_curve.png', names, ylabel='Precision')
plot_mc_curve(px, r, Path(save_dir) / f'{prefix}R_curve.png', names, ylabel='Recall')
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')
plot_mc_curve(px, p, Path(save_dir) / 'P_curve.png', names, ylabel='Precision')
plot_mc_curve(px, r, Path(save_dir) / 'R_curve.png', names, ylabel='Recall')
i = smooth(f1.mean(0), 0.1).argmax() # max F1 index
p, r, f1 = p[:, i], r[:, i], f1[:, i]
tp = (r * nt).round() # true positives
fp = (tp / (p + eps) - tp).round() # false positives
return tp, fp, p, r, f1, ap, unique_classes.astype(int)
i = f1.mean(0).argmax() # max F1 index
return p[:, i], r[:, i], ap, f1[:, i], unique_classes.astype('int32')
def compute_ap(recall, precision):
@@ -141,12 +129,6 @@ class ConfusionMatrix:
Returns:
None, updates confusion matrix accordingly
"""
if detections is None:
gt_classes = labels.int()
for gc in gt_classes:
self.matrix[self.nc, gc] += 1 # background FN
return
detections = detections[detections[:, 4] > self.conf]
gt_classes = labels[:, 0].int()
detection_classes = detections[:, 5].int()
@@ -164,55 +146,43 @@ class ConfusionMatrix:
matches = np.zeros((0, 3))
n = matches.shape[0] > 0
m0, m1, _ = matches.transpose().astype(int)
m0, m1, _ = matches.transpose().astype(np.int16)
for i, gc in enumerate(gt_classes):
j = m0 == i
if n and sum(j) == 1:
self.matrix[detection_classes[m1[j]], gc] += 1 # correct
else:
self.matrix[self.nc, gc] += 1 # true background
self.matrix[self.nc, gc] += 1 # background FP
if n:
for i, dc in enumerate(detection_classes):
if not any(m1 == i):
self.matrix[dc, self.nc] += 1 # predicted background
self.matrix[dc, self.nc] += 1 # background FN
def tp_fp(self):
tp = self.matrix.diagonal() # true positives
fp = self.matrix.sum(1) - tp # false positives
# fn = self.matrix.sum(0) - tp # false negatives (missed detections)
return tp[:-1], fp[:-1] # remove background class
def matrix(self):
return self.matrix
@TryExcept('WARNING ⚠️ ConfusionMatrix plot failure')
def plot(self, normalize=True, save_dir='', names=()):
import seaborn as sn
try:
import seaborn as sn
array = self.matrix / ((self.matrix.sum(0).reshape(1, -1) + 1E-9) if normalize else 1) # normalize columns
array[array < 0.005] = np.nan # don't annotate (would appear as 0.00)
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, ax = plt.subplots(1, 1, figsize=(12, 9), tight_layout=True)
nc, nn = self.nc, len(names) # number of classes, names
sn.set(font_scale=1.0 if nc < 50 else 0.8) # for label size
labels = (0 < nn < 99) and (nn == nc) # apply names to ticklabels
ticklabels = (names + ['background']) if labels else 'auto'
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress empty matrix RuntimeWarning: All-NaN slice encountered
sn.heatmap(array,
ax=ax,
annot=nc < 30,
annot_kws={
'size': 8},
cmap='Blues',
fmt='.2f',
square=True,
vmin=0.0,
xticklabels=ticklabels,
yticklabels=ticklabels).set_facecolor((1, 1, 1))
ax.set_xlabel('True')
ax.set_ylabel('Predicted')
ax.set_title('Confusion Matrix')
fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250)
plt.close(fig)
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
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:
print(f'WARNING: ConfusionMatrix plot failure: {e}')
def print(self):
for i in range(self.nc + 1):
@@ -224,19 +194,19 @@ def bbox_iou(box1, box2, xywh=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7
# Get the coordinates of bounding boxes
if xywh: # transform from xywh to xyxy
(x1, y1, w1, h1), (x2, y2, w2, h2) = box1.chunk(4, -1), box2.chunk(4, -1)
(x1, y1, w1, h1), (x2, y2, w2, h2) = box1.chunk(4, 1), box2.chunk(4, 1)
w1_, h1_, w2_, h2_ = w1 / 2, h1 / 2, w2 / 2, h2 / 2
b1_x1, b1_x2, b1_y1, b1_y2 = x1 - w1_, x1 + w1_, y1 - h1_, y1 + h1_
b2_x1, b2_x2, b2_y1, b2_y2 = x2 - w2_, x2 + w2_, y2 - h2_, y2 + h2_
else: # x1, y1, x2, y2 = box1
b1_x1, b1_y1, b1_x2, b1_y2 = box1.chunk(4, -1)
b2_x1, b2_y1, b2_x2, b2_y2 = box2.chunk(4, -1)
w1, h1 = b1_x2 - b1_x1, (b1_y2 - b1_y1).clamp(eps)
w2, h2 = b2_x2 - b2_x1, (b2_y2 - b2_y1).clamp(eps)
b1_x1, b1_y1, b1_x2, b1_y2 = box1.chunk(4, 1)
b2_x1, b2_y1, b2_x2, b2_y2 = box2.chunk(4, 1)
w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps
# Intersection area
inter = (b1_x2.minimum(b2_x2) - b1_x1.maximum(b2_x1)).clamp(0) * \
(b1_y2.minimum(b2_y2) - b1_y1.maximum(b2_y1)).clamp(0)
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
union = w1 * h1 + w2 * h2 - inter + eps
@@ -244,13 +214,13 @@ def bbox_iou(box1, box2, xywh=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7
# IoU
iou = inter / union
if CIoU or DIoU or GIoU:
cw = b1_x2.maximum(b2_x2) - b1_x1.minimum(b2_x1) # convex (smallest enclosing box) width
ch = b1_y2.maximum(b2_y2) - b1_y1.minimum(b2_y1) # convex height
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 dist ** 2
if CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
v = (4 / math.pi ** 2) * (torch.atan(w2 / h2) - torch.atan(w1 / h1)).pow(2)
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
@@ -260,7 +230,7 @@ def bbox_iou(box1, box2, xywh=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7
return iou # IoU
def box_iou(box1, box2, eps=1e-7):
def box_iou(box1, box2):
# https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
"""
Return intersection-over-union (Jaccard index) of boxes.
@@ -273,24 +243,30 @@ def box_iou(box1, box2, eps=1e-7):
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)
(a1, a2), (b1, b2) = box1.unsqueeze(1).chunk(2, 2), box2.unsqueeze(0).chunk(2, 2)
inter = (torch.min(a2, b2) - torch.max(a1, b1)).clamp(0).prod(2)
# IoU = inter / (area1 + area2 - inter)
return inter / ((a2 - a1).prod(2) + (b2 - b1).prod(2) - inter + eps)
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):
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
b2_x1, b2_y1, b2_x2, b2_y2 = box2.T
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) * \
@@ -303,19 +279,17 @@ def bbox_ioa(box1, box2, eps=1e-7):
return inter_area / box2_area
def wh_iou(wh1, wh2, eps=1e-7):
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 + eps) # iou = inter / (area1 + area2 - inter)
return inter / (wh1.prod(2) + wh2.prod(2) - inter) # iou = inter / (area1 + area2 - inter)
# Plots ----------------------------------------------------------------------------------------------------------------
@threaded
def plot_pr_curve(px, py, ap, save_dir=Path('pr_curve.png'), names=()):
def plot_pr_curve(px, py, ap, save_dir='pr_curve.png', names=()):
# Precision-recall curve
fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
py = np.stack(py, axis=1)
@@ -331,14 +305,12 @@ def plot_pr_curve(px, py, ap, save_dir=Path('pr_curve.png'), names=()):
ax.set_ylabel('Precision')
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
ax.legend(bbox_to_anchor=(1.04, 1), loc='upper left')
ax.set_title('Precision-Recall Curve')
fig.savefig(save_dir, dpi=250)
plt.close(fig)
plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
fig.savefig(Path(save_dir), dpi=250)
plt.close()
@threaded
def plot_mc_curve(px, py, save_dir=Path('mc_curve.png'), names=(), xlabel='Confidence', ylabel='Metric'):
def plot_mc_curve(px, py, save_dir='mc_curve.png', names=(), xlabel='Confidence', ylabel='Metric'):
# Metric-confidence curve
fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
@@ -348,13 +320,12 @@ def plot_mc_curve(px, py, save_dir=Path('mc_curve.png'), names=(), xlabel='Confi
else:
ax.plot(px, py.T, linewidth=1, color='grey') # plot(confidence, metric)
y = smooth(py.mean(0), 0.05)
y = py.mean(0)
ax.plot(px, y, linewidth=3, color='blue', label=f'all classes {y.max():.2f} at {px[y.argmax()]:.3f}')
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
ax.legend(bbox_to_anchor=(1.04, 1), loc='upper left')
ax.set_title(f'{ylabel}-Confidence Curve')
fig.savefig(save_dir, dpi=250)
plt.close(fig)
plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
fig.savefig(Path(save_dir), dpi=250)
plt.close()
yolov3/utils/plots.py
+55 -146
View File
@@ -3,12 +3,10 @@
Plotting utils
"""
import contextlib
import math
import os
from copy import copy
from pathlib import Path
from urllib.error import URLError
import cv2
import matplotlib
@@ -19,13 +17,12 @@ import seaborn as sn
import torch
from PIL import Image, ImageDraw, ImageFont
from utils import TryExcept, threaded
from utils.general import (CONFIG_DIR, FONT, LOGGER, check_font, check_requirements, clip_boxes, increment_path,
is_ascii, 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
from utils.segment.general import scale_image
# 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
@@ -35,9 +32,9 @@ class Colors:
# Ultralytics color palette https://ultralytics.com/
def __init__(self):
# hex = matplotlib.colors.TABLEAU_COLORS.values()
hexs = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
'2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
self.palette = [self.hex2rgb(f'#{c}') for c in hexs]
hex = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
'2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
self.palette = [self.hex2rgb('#' + c) for c in hex]
self.n = len(self.palette)
def __call__(self, i, bgr=False):
@@ -52,33 +49,35 @@ class Colors:
colors = Colors() # create instance for 'from utils.plots import colors'
def check_pil_font(font=FONT, size=10):
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: # download if missing
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:
check_font(font)
return ImageFont.truetype(str(font), size)
except TypeError:
check_requirements('Pillow>=8.4.0') # known issue https://github.com/ultralytics/yolov5/issues/5374
except URLError: # not online
return ImageFont.load_default()
class Annotator:
# YOLOv3 Annotator for train/val mosaics and jpgs and detect/hub inference annotations
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.'
non_ascii = not is_ascii(example) # non-latin labels, i.e. asian, arabic, cyrillic
self.pil = pil or non_ascii
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_pil_font(font='Arial.Unicode.ttf' if non_ascii else font,
size=font_size or max(round(sum(self.im.size) / 2 * 0.035), 12))
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
@@ -88,14 +87,12 @@ class Annotator:
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 (WARNING: deprecated) in 9.2.0
# _, _, w, h = self.font.getbbox(label) # text width, height (New)
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.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
@@ -104,62 +101,20 @@ class Annotator:
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
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 masks(self, masks, colors, im_gpu, alpha=0.5, retina_masks=False):
"""Plot masks at once.
Args:
masks (tensor): predicted masks on cuda, shape: [n, h, w]
colors (List[List[Int]]): colors for predicted masks, [[r, g, b] * n]
im_gpu (tensor): img is in cuda, shape: [3, h, w], range: [0, 1]
alpha (float): mask transparency: 0.0 fully transparent, 1.0 opaque
"""
if self.pil:
# convert to numpy first
self.im = np.asarray(self.im).copy()
if len(masks) == 0:
self.im[:] = im_gpu.permute(1, 2, 0).contiguous().cpu().numpy() * 255
colors = torch.tensor(colors, device=im_gpu.device, dtype=torch.float32) / 255.0
colors = colors[:, None, None] # shape(n,1,1,3)
masks = masks.unsqueeze(3) # shape(n,h,w,1)
masks_color = masks * (colors * alpha) # shape(n,h,w,3)
inv_alph_masks = (1 - masks * alpha).cumprod(0) # shape(n,h,w,1)
mcs = (masks_color * inv_alph_masks).sum(0) * 2 # mask color summand shape(n,h,w,3)
im_gpu = im_gpu.flip(dims=[0]) # flip channel
im_gpu = im_gpu.permute(1, 2, 0).contiguous() # shape(h,w,3)
im_gpu = im_gpu * inv_alph_masks[-1] + mcs
im_mask = (im_gpu * 255).byte().cpu().numpy()
self.im[:] = im_mask if retina_masks else scale_image(im_gpu.shape, im_mask, self.im.shape)
if self.pil:
# convert im back to PIL and update draw
self.fromarray(self.im)
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), anchor='top'):
def text(self, xy, text, txt_color=(255, 255, 255)):
# Add text to image (PIL-only)
if anchor == 'bottom': # start y from font bottom
w, h = self.font.getsize(text) # text width, height
xy[1] += 1 - h
self.draw.text(xy, text, fill=txt_color, font=self.font)
def fromarray(self, im):
# Update self.im from a numpy array
self.im = im if isinstance(im, Image.Image) else Image.fromarray(im)
self.draw = ImageDraw.Draw(self.im)
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
@@ -177,7 +132,7 @@ def feature_visualization(x, module_type, stage, n=32, save_dir=Path('runs/detec
if 'Detect' not in module_type:
batch, channels, height, width = x.shape # batch, channels, height, width
if height > 1 and width > 1:
f = save_dir / f"stage{stage}_{module_type.split('.')[-1]}_features.png" # filename
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
@@ -188,10 +143,9 @@ def feature_visualization(x, module_type, stage, n=32, save_dir=Path('runs/detec
ax[i].imshow(blocks[i].squeeze()) # cmap='gray'
ax[i].axis('off')
LOGGER.info(f'Saving {f}... ({n}/{channels})')
plt.savefig(f, dpi=300, bbox_inches='tight')
print(f'Saving {save_dir / f}... ({n}/{channels})')
plt.savefig(save_dir / f, dpi=300, bbox_inches='tight')
plt.close()
np.save(str(f.with_suffix('.npy')), x[0].cpu().numpy()) # npy save
def hist2d(x, y, n=100):
@@ -216,31 +170,26 @@ def butter_lowpass_filtfilt(data, cutoff=1500, fs=50000, order=5):
return filtfilt(b, a, data) # forward-backward filter
def output_to_target(output, max_det=300):
# Convert model output to target format [batch_id, class_id, x, y, w, h, conf] for plotting
def output_to_target(output):
# Convert model output to target format [batch_id, class_id, x, y, w, h, conf]
targets = []
for i, o in enumerate(output):
box, conf, cls = o[:max_det, :6].cpu().split((4, 1, 1), 1)
j = torch.full((conf.shape[0], 1), i)
targets.append(torch.cat((j, cls, xyxy2xywh(box), conf), 1))
return torch.cat(targets, 0).numpy()
for *box, conf, cls in o.cpu().numpy():
targets.append([i, cls, *list(*xyxy2xywh(np.array(box)[None])), conf])
return np.array(targets)
@threaded
def plot_images(images, targets, paths=None, fname='images.jpg', names=None):
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()
max_size = 1920 # max image size
max_subplots = 16 # max image subplots, i.e. 4x4
if np.max(images[0]) <= 1:
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)
if np.max(images[0]) <= 1:
images *= 255 # de-normalise (optional)
# Build Image
mosaic = np.full((int(ns * h), int(ns * w), 3), 255, dtype=np.uint8) # init
@@ -260,12 +209,12 @@ def plot_images(images, targets, paths=None, fname='images.jpg', names=None):
# Annotate
fs = int((h + w) * ns * 0.01) # font size
annotator = Annotator(mosaic, line_width=round(fs / 10), font_size=fs, pil=True, example=names)
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), text=Path(paths[i]).name[:40], txt_color=(220, 220, 220)) # filenames
annotator.text((x + 5, y + 5 + h), text=Path(paths[i]).name[:40], txt_color=(220, 220, 220)) # filenames
if len(targets) > 0:
ti = targets[targets[:, 0] == i] # image targets
boxes = xywh2xyxy(ti[:, 2:6]).T
@@ -346,7 +295,7 @@ def plot_val_study(file='', dir='', x=None): # from utils.plots import *; plot_
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 [save_dir / f'study_coco_{x}.txt' for x in ['yolov5n6', 'yolov5s6', 'yolov5m6', 'yolov5l6', 'yolov5x6']]:
# 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)
@@ -357,19 +306,11 @@ def plot_val_study(file='', dir='', x=None): # from utils.plots import *; plot_
ax[i].set_title(s[i])
j = y[3].argmax() + 1
ax2.plot(y[5, 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],
'k.-',
linewidth=2,
markersize=8,
alpha=.25,
label='EfficientDet')
'k.-', linewidth=2, markersize=8, alpha=.25, label='EfficientDet')
ax2.grid(alpha=0.2)
ax2.set_yticks(np.arange(20, 60, 5))
@@ -383,7 +324,8 @@ def plot_val_study(file='', dir='', x=None): # from utils.plots import *; plot_
plt.savefig(f, dpi=300)
@TryExcept() # known issue https://github.com/ultralytics/yolov5/issues/5395
@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
LOGGER.info(f"Plotting labels to {save_dir / 'labels.jpg'}... ")
@@ -400,12 +342,11 @@ def plot_labels(labels, names=(), save_dir=Path('')):
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)
with contextlib.suppress(Exception): # color histogram bars by class
[y[2].patches[i].set_color([x / 255 for x in colors(i)]) for i in range(nc)] # known issue #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(list(names.values()), rotation=90, fontsize=10)
ax[0].set_xticklabels(names, rotation=90, fontsize=10)
else:
ax[0].set_xlabel('classes')
sn.histplot(x, x='x', y='y', ax=ax[2], bins=50, pmax=0.9)
@@ -429,35 +370,6 @@ def plot_labels(labels, names=(), save_dir=Path('')):
plt.close()
def imshow_cls(im, labels=None, pred=None, names=None, nmax=25, verbose=False, f=Path('images.jpg')):
# Show classification image grid with labels (optional) and predictions (optional)
from utils.augmentations import denormalize
names = names or [f'class{i}' for i in range(1000)]
blocks = torch.chunk(denormalize(im.clone()).cpu().float(), len(im),
dim=0) # select batch index 0, block by channels
n = min(len(blocks), nmax) # number of plots
m = min(8, round(n ** 0.5)) # 8 x 8 default
fig, ax = plt.subplots(math.ceil(n / m), m) # 8 rows x n/8 cols
ax = ax.ravel() if m > 1 else [ax]
# plt.subplots_adjust(wspace=0.05, hspace=0.05)
for i in range(n):
ax[i].imshow(blocks[i].squeeze().permute((1, 2, 0)).numpy().clip(0.0, 1.0))
ax[i].axis('off')
if labels is not None:
s = names[labels[i]] + (f'—{names[pred[i]]}' if pred is not None else '')
ax[i].set_title(s, fontsize=8, verticalalignment='top')
plt.savefig(f, dpi=300, bbox_inches='tight')
plt.close()
if verbose:
LOGGER.info(f'Saving {f}')
if labels is not None:
LOGGER.info('True: ' + ' '.join(f'{names[i]:3s}' for i in labels[:nmax]))
if pred is not None:
LOGGER.info('Predicted:' + ' '.join(f'{names[i]:3s}' for i in pred[:nmax]))
return f
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)
@@ -468,7 +380,6 @@ def plot_evolve(evolve_csv='path/to/evolve.csv'): # from utils.plots import *;
j = np.argmax(f) # max fitness index
plt.figure(figsize=(10, 12), tight_layout=True)
matplotlib.rc('font', **{'size': 8})
print(f'Best results from row {j} of {evolve_csv}:')
for i, k in enumerate(keys[7:]):
v = x[:, 7 + i]
mu = v[j] # best single result
@@ -492,20 +403,20 @@ def plot_results(file='path/to/results.csv', dir=''):
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 f in files:
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].astype('float')
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:
LOGGER.info(f'Warning: Plotting error for {f}: {e}')
print(f'Warning: Plotting error for {f}: {e}')
ax[1].legend()
fig.savefig(save_dir / 'results.png', dpi=200)
plt.close()
@@ -542,7 +453,7 @@ def profile_idetection(start=0, stop=0, labels=(), save_dir=''):
plt.savefig(Path(save_dir) / 'idetection_profile.png', dpi=200)
def save_one_box(xyxy, im, file=Path('im.jpg'), gain=1.02, pad=10, square=False, BGR=False, save=True):
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
@@ -550,11 +461,9 @@ def save_one_box(xyxy, im, file=Path('im.jpg'), gain=1.02, pad=10, square=False,
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_boxes(xyxy, im.shape)
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
f = str(increment_path(file).with_suffix('.jpg'))
# cv2.imwrite(f, crop) # save BGR, https://github.com/ultralytics/yolov5/issues/7007 chroma subsampling issue
Image.fromarray(crop[..., ::-1]).save(f, quality=95, subsampling=0) # save RGB
cv2.imwrite(str(increment_path(file).with_suffix('.jpg')), crop)
return crop
yolov3/utils/torch_utils.py
+89 -203
View File
@@ -3,12 +3,12 @@
PyTorch utils
"""
import datetime
import math
import os
import platform
import subprocess
import time
import warnings
from contextlib import contextmanager
from copy import deepcopy
from pathlib import Path
@@ -17,77 +17,20 @@ import torch
import torch.distributed as dist
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.parallel import DistributedDataParallel as DDP
from utils.general import LOGGER, check_version, colorstr, file_date, git_describe
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
from utils.general import LOGGER
try:
import thop # for FLOPs computation
except ImportError:
thop = None
# Suppress PyTorch warnings
warnings.filterwarnings('ignore', message='User provided device_type of \'cuda\', but CUDA is not available. Disabling')
warnings.filterwarnings('ignore', category=UserWarning)
def smart_inference_mode(torch_1_9=check_version(torch.__version__, '1.9.0')):
# Applies torch.inference_mode() decorator if torch>=1.9.0 else torch.no_grad() decorator
def decorate(fn):
return (torch.inference_mode if torch_1_9 else torch.no_grad)()(fn)
return decorate
def smartCrossEntropyLoss(label_smoothing=0.0):
# Returns nn.CrossEntropyLoss with label smoothing enabled for torch>=1.10.0
if check_version(torch.__version__, '1.10.0'):
return nn.CrossEntropyLoss(label_smoothing=label_smoothing)
if label_smoothing > 0:
LOGGER.warning(f'WARNING ⚠️ label smoothing {label_smoothing} requires torch>=1.10.0')
return nn.CrossEntropyLoss()
def smart_DDP(model):
# Model DDP creation with checks
assert not check_version(torch.__version__, '1.12.0', pinned=True), \
'torch==1.12.0 torchvision==0.13.0 DDP training is not supported due to a known issue. ' \
'Please upgrade or downgrade torch to use DDP. See https://github.com/ultralytics/yolov5/issues/8395'
if check_version(torch.__version__, '1.11.0'):
return DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK, static_graph=True)
else:
return DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK)
def reshape_classifier_output(model, n=1000):
# Update a TorchVision classification model to class count 'n' if required
from models.common import Classify
name, m = list((model.model if hasattr(model, 'model') else model).named_children())[-1] # last module
if isinstance(m, Classify): # YOLOv3 Classify() head
if m.linear.out_features != n:
m.linear = nn.Linear(m.linear.in_features, n)
elif isinstance(m, nn.Linear): # ResNet, EfficientNet
if m.out_features != n:
setattr(model, name, nn.Linear(m.in_features, n))
elif isinstance(m, nn.Sequential):
types = [type(x) for x in m]
if nn.Linear in types:
i = types.index(nn.Linear) # nn.Linear index
if m[i].out_features != n:
m[i] = nn.Linear(m[i].in_features, n)
elif nn.Conv2d in types:
i = types.index(nn.Conv2d) # nn.Conv2d index
if m[i].out_channels != n:
m[i] = nn.Conv2d(m[i].in_channels, n, m[i].kernel_size, m[i].stride, bias=m[i].bias is not None)
@contextmanager
def torch_distributed_zero_first(local_rank: int):
# Decorator to make all processes in distributed training wait for each local_master to do something
"""
Decorator to make all processes in distributed training wait for each local_master to do something.
"""
if local_rank not in [-1, 0]:
dist.barrier(device_ids=[local_rank])
yield
@@ -95,70 +38,69 @@ def torch_distributed_zero_first(local_rank: int):
dist.barrier(device_ids=[0])
def device_count():
# Returns number of CUDA devices available. Safe version of torch.cuda.device_count(). Supports Linux and Windows
assert platform.system() in ('Linux', 'Windows'), 'device_count() only supported on Linux or Windows'
def date_modified(path=__file__):
# return human-readable file modification date, i.e. '2021-3-26'
t = datetime.datetime.fromtimestamp(Path(path).stat().st_mtime)
return f'{t.year}-{t.month}-{t.day}'
def git_describe(path=Path(__file__).parent): # path must be a directory
# return human-readable git description, i.e. v5.0-5-g3e25f1e https://git-scm.com/docs/git-describe
s = f'git -C {path} describe --tags --long --always'
try:
cmd = 'nvidia-smi -L | wc -l' if platform.system() == 'Linux' else 'nvidia-smi -L | find /c /v ""' # Windows
return int(subprocess.run(cmd, shell=True, capture_output=True, check=True).stdout.decode().split()[-1])
except Exception:
return 0
return subprocess.check_output(s, shell=True, stderr=subprocess.STDOUT).decode()[:-1]
except subprocess.CalledProcessError as e:
return '' # not a git repository
def select_device(device='', batch_size=0, newline=True):
# device = None or 'cpu' or 0 or '0' or '0,1,2,3'
s = f'YOLOv3 🚀 {git_describe() or file_date()} Python-{platform.python_version()} torch-{torch.__version__} '
device = str(device).strip().lower().replace('cuda:', '').replace('none', '') # to string, 'cuda:0' to '0'
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
device = str(device).strip().lower().replace('cuda:', '') # to string, 'cuda:0' to '0'
cpu = device == 'cpu'
mps = device == 'mps' # Apple Metal Performance Shaders (MPS)
if cpu or mps:
if cpu:
os.environ['CUDA_VISIBLE_DEVICES'] = '-1' # force torch.cuda.is_available() = False
elif device: # non-cpu device requested
os.environ['CUDA_VISIBLE_DEVICES'] = device # set environment variable - must be before assert is_available()
assert torch.cuda.is_available() and torch.cuda.device_count() >= len(device.replace(',', '')), \
f"Invalid CUDA '--device {device}' requested, use '--device cpu' or pass valid CUDA device(s)"
os.environ['CUDA_VISIBLE_DEVICES'] = device # set environment variable
assert torch.cuda.is_available(), f'CUDA unavailable, invalid device {device} requested' # check availability
if not cpu and not mps and torch.cuda.is_available(): # prefer GPU if available
cuda = not cpu and torch.cuda.is_available()
if cuda:
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 > 0: # check batch_size is divisible by 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) + 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 / (1 << 20):.0f}MiB)\n" # bytes to MB
arg = 'cuda:0'
elif mps and getattr(torch, 'has_mps', False) and torch.backends.mps.is_available(): # prefer MPS if available
s += 'MPS\n'
arg = 'mps'
else: # revert to CPU
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'
arg = 'cpu'
if not newline:
s = s.rstrip()
LOGGER.info(s)
return torch.device(arg)
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_sync():
# PyTorch-accurate time
# pytorch-accurate time
if torch.cuda.is_available():
torch.cuda.synchronize()
return time.time()
def profile(input, ops, n=10, device=None):
""" YOLOv3 speed/memory/FLOPs profiler
Usage:
input = torch.randn(16, 3, 640, 640)
m1 = lambda x: x * torch.sigmoid(x)
m2 = nn.SiLU()
profile(input, [m1, m2], n=100) # profile over 100 iterations
"""
# speed/memory/FLOPs profiler
#
# Usage:
# input = torch.randn(16, 3, 640, 640)
# m1 = lambda x: x * torch.sigmoid(x)
# m2 = nn.SiLU()
# profile(input, [m1, m2], n=100) # profile over 100 iterations
results = []
if not isinstance(device, torch.device):
device = select_device(device)
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}")
@@ -171,7 +113,7 @@ def profile(input, ops, n=10, device=None):
tf, tb, t = 0, 0, [0, 0, 0] # dt forward, backward
try:
flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # GFLOPs
except Exception:
except:
flops = 0
try:
@@ -182,14 +124,15 @@ def profile(input, ops, n=10, device=None):
try:
_ = (sum(yi.sum() for yi in y) if isinstance(y, list) else y).sum().backward()
t[2] = time_sync()
except Exception: # no backward method
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, s_out = (tuple(x.shape) if isinstance(x, torch.Tensor) else 'list' for x in (x, y)) # shapes
p = sum(x.numel() for x in m.parameters()) if isinstance(m, nn.Module) else 0 # parameters
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:
@@ -238,30 +181,30 @@ def sparsity(model):
def prune(model, amount=0.3):
# Prune model to requested global sparsity
import torch.nn.utils.prune as prune
print('Pruning model... ', end='')
for name, m in model.named_modules():
if isinstance(m, nn.Conv2d):
prune.l1_unstructured(m, name='weight', amount=amount) # prune
prune.remove(m, 'weight') # make permanent
LOGGER.info(f'Model pruned to {sparsity(model):.3g} global sparsity')
print(' %.3g global sparsity' % sparsity(model))
def fuse_conv_and_bn(conv, bn):
# Fuse Conv2d() and BatchNorm2d() layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/
# Fuse convolution and batchnorm layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/
fusedconv = nn.Conv2d(conv.in_channels,
conv.out_channels,
kernel_size=conv.kernel_size,
stride=conv.stride,
padding=conv.padding,
dilation=conv.dilation,
groups=conv.groups,
bias=True).requires_grad_(False).to(conv.weight.device)
# Prepare filters
# prepare filters
w_conv = conv.weight.clone().view(conv.out_channels, -1)
w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape))
# Prepare spatial bias
# prepare spatial bias
b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias
b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
@@ -269,7 +212,7 @@ def fuse_conv_and_bn(conv, bn):
return fusedconv
def model_info(model, verbose=False, imgsz=640):
def model_info(model, verbose=False, img_size=640):
# Model information. img_size may be int or list, i.e. img_size=640 or img_size=[640, 320]
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
@@ -281,29 +224,29 @@ def model_info(model, verbose=False, imgsz=640):
(i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
try: # FLOPs
p = next(model.parameters())
stride = max(int(model.stride.max()), 32) if hasattr(model, 'stride') else 32 # max stride
im = torch.empty((1, p.shape[1], stride, stride), device=p.device) # input image in BCHW format
flops = thop.profile(deepcopy(model), inputs=(im,), verbose=False)[0] / 1E9 * 2 # stride GFLOPs
imgsz = imgsz if isinstance(imgsz, list) else [imgsz, imgsz] # expand if int/float
fs = f', {flops * imgsz[0] / stride * imgsz[1] / stride:.1f} GFLOPs' # 640x640 GFLOPs
except Exception:
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
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
except (ImportError, Exception):
fs = ''
name = Path(model.yaml_file).stem.replace('yolov5', 'YOLOv3') if hasattr(model, 'yaml_file') else 'Model'
LOGGER.info(f'{name} summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}')
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)
# Scales img(bs,3,y,x) by ratio constrained to gs-multiple
# scales img(bs,3,y,x) by ratio constrained to gs-multiple
if ratio == 1.0:
return img
h, w = img.shape[2:]
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))
return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447) # value = imagenet mean
else:
h, w = img.shape[2:]
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))
return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447) # value = imagenet mean
def copy_attr(a, b, include=(), exclude=()):
@@ -315,71 +258,8 @@ def copy_attr(a, b, include=(), exclude=()):
setattr(a, k, v)
def smart_optimizer(model, name='Adam', lr=0.001, momentum=0.9, decay=1e-5):
# YOLOv3 3-param group optimizer: 0) weights with decay, 1) weights no decay, 2) biases no decay
g = [], [], [] # optimizer parameter groups
bn = tuple(v for k, v in nn.__dict__.items() if 'Norm' in k) # normalization layers, i.e. BatchNorm2d()
for v in model.modules():
for p_name, p in v.named_parameters(recurse=0):
if p_name == 'bias': # bias (no decay)
g[2].append(p)
elif p_name == 'weight' and isinstance(v, bn): # weight (no decay)
g[1].append(p)
else:
g[0].append(p) # weight (with decay)
if name == 'Adam':
optimizer = torch.optim.Adam(g[2], lr=lr, betas=(momentum, 0.999)) # adjust beta1 to momentum
elif name == 'AdamW':
optimizer = torch.optim.AdamW(g[2], lr=lr, betas=(momentum, 0.999), weight_decay=0.0)
elif name == 'RMSProp':
optimizer = torch.optim.RMSprop(g[2], lr=lr, momentum=momentum)
elif name == 'SGD':
optimizer = torch.optim.SGD(g[2], lr=lr, momentum=momentum, nesterov=True)
else:
raise NotImplementedError(f'Optimizer {name} not implemented.')
optimizer.add_param_group({'params': g[0], 'weight_decay': decay}) # add g0 with weight_decay
optimizer.add_param_group({'params': g[1], 'weight_decay': 0.0}) # add g1 (BatchNorm2d weights)
LOGGER.info(f"{colorstr('optimizer:')} {type(optimizer).__name__}(lr={lr}) with parameter groups "
f'{len(g[1])} weight(decay=0.0), {len(g[0])} weight(decay={decay}), {len(g[2])} bias')
return optimizer
def smart_hub_load(repo='ultralytics/yolov5', model='yolov5s', **kwargs):
# YOLOv3 torch.hub.load() wrapper with smart error/issue handling
if check_version(torch.__version__, '1.9.1'):
kwargs['skip_validation'] = True # validation causes GitHub API rate limit errors
if check_version(torch.__version__, '1.12.0'):
kwargs['trust_repo'] = True # argument required starting in torch 0.12
try:
return torch.hub.load(repo, model, **kwargs)
except Exception:
return torch.hub.load(repo, model, force_reload=True, **kwargs)
def smart_resume(ckpt, optimizer, ema=None, weights='yolov5s.pt', epochs=300, resume=True):
# Resume training from a partially trained checkpoint
best_fitness = 0.0
start_epoch = ckpt['epoch'] + 1
if ckpt['optimizer'] is not None:
optimizer.load_state_dict(ckpt['optimizer']) # optimizer
best_fitness = ckpt['best_fitness']
if ema and ckpt.get('ema'):
ema.ema.load_state_dict(ckpt['ema'].float().state_dict()) # EMA
ema.updates = ckpt['updates']
if resume:
assert start_epoch > 0, f'{weights} training to {epochs} epochs is finished, nothing to resume.\n' \
f"Start a new training without --resume, i.e. 'python train.py --weights {weights}'"
LOGGER.info(f'Resuming training from {weights} from epoch {start_epoch} to {epochs} total epochs')
if epochs < start_epoch:
LOGGER.info(f"{weights} has been trained for {ckpt['epoch']} epochs. Fine-tuning for {epochs} more epochs.")
epochs += ckpt['epoch'] # finetune additional epochs
return best_fitness, start_epoch, epochs
class EarlyStopping:
# YOLOv3 simple early stopper
# simple early stopper
def __init__(self, patience=30):
self.best_fitness = 0.0 # i.e. mAP
self.best_epoch = 0
@@ -402,30 +282,36 @@ class EarlyStopping:
class ModelEMA:
""" Updated Exponential Moving Average (EMA) from https://github.com/rwightman/pytorch-image-models
Keeps a moving average of everything in the model state_dict (parameters and buffers)
For EMA details see https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
""" 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).
This is intended to allow functionality like
https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
A smoothed version of the weights is necessary for some training schemes to perform well.
This class is sensitive where it is initialized in the sequence of model init,
GPU assignment and distributed training wrappers.
"""
def __init__(self, model, decay=0.9999, tau=2000, updates=0):
def __init__(self, model, decay=0.9999, updates=0):
# Create EMA
self.ema = deepcopy(de_parallel(model)).eval() # FP32 EMA
self.ema = deepcopy(model.module if is_parallel(model) else model).eval() # FP32 EMA
# if next(model.parameters()).device.type != 'cpu':
# self.ema.half() # FP16 EMA
self.updates = updates # number of EMA updates
self.decay = lambda x: decay * (1 - math.exp(-x / tau)) # decay exponential ramp (to help early epochs)
self.decay = lambda x: decay * (1 - math.exp(-x / 2000)) # decay exponential ramp (to help early epochs)
for p in self.ema.parameters():
p.requires_grad_(False)
def update(self, model):
# Update EMA parameters
self.updates += 1
d = self.decay(self.updates)
with torch.no_grad():
self.updates += 1
d = self.decay(self.updates)
msd = de_parallel(model).state_dict() # model state_dict
for k, v in self.ema.state_dict().items():
if v.dtype.is_floating_point: # true for FP16 and FP32
v *= d
v += (1 - d) * msd[k].detach()
# assert v.dtype == msd[k].dtype == torch.float32, f'{k}: EMA {v.dtype} and model {msd[k].dtype} must be FP32'
msd = model.module.state_dict() if is_parallel(model) else model.state_dict() # model state_dict
for k, v in self.ema.state_dict().items():
if v.dtype.is_floating_point:
v *= d
v += (1 - d) * msd[k].detach()
def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
# Update EMA attributes