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

YOLOv5 Forward Compatibility Update (#1569)

Browse Source 
* YOLOv5 forward compatibility update

* add data dir

* ci test yolov3

* update build_targets()

* update build_targets()

* update build_targets()

* update yolov3-spp.yaml

* add yolov3-tiny.yaml

* add yolov3-tiny.yaml

* Update yolov3-tiny.yaml

* thop bug fix

* Detection() device bug fix

* Use torchvision.ops.nms()

* Remove redundant download mirror

* CI tests with yolov3-tiny

* Update README.md

* Synch train and test iou_thresh

* update requirements.txt

* Cat apriori autolabels

* Confusion matrix

* Autosplit

* Autosplit

* Update README.md

* AP no plot

* Update caching

* Update caching

* Caching bug fix

* --image-weights bug fix

* datasets bug fix

* mosaic plots bug fix

* plot_study

* boxes.max()

* boxes.max()

* boxes.max()

* boxes.max()

* boxes.max()

* boxes.max()

* update

* Update README

* Update README

* Update README.md

* Update README.md

* results png

* Update README

* Targets scaling bug fix

* update plot_study

* update plot_study

* update plot_study

* update plot_study

* Targets scaling bug fix

* Finish Readme.md

* Finish Readme.md

* Finish Readme.md

* Update README.md

* Creado con Colaboratory
This commit is contained in:
Glenn Jocher
2020-11-26 20:24:00 +01:00
committed by GitHub
parent 98068efebc
commit 76807fae71
87 changed files with 5613 additions and 18673 deletions
Download Patch File Download Diff File Expand all files Collapse all files
utils/activations.py
+72
View File
@@ -0,0 +1,72 @@
# Activation functions
import torch
import torch.nn as nn
import torch.nn.functional as F
# Swish https://arxiv.org/pdf/1905.02244.pdf ---------------------------------------------------------------------------
class Swish(nn.Module): #
@staticmethod
def forward(x):
return x * torch.sigmoid(x)
class Hardswish(nn.Module): # export-friendly version of nn.Hardswish()
@staticmethod
def forward(x):
# return x * F.hardsigmoid(x) # for torchscript and CoreML
return x * F.hardtanh(x + 3, 0., 6.) / 6. # for torchscript, CoreML and ONNX
class MemoryEfficientSwish(nn.Module):
class F(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
return x * torch.sigmoid(x)
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_tensors[0]
sx = torch.sigmoid(x)
return grad_output * (sx * (1 + x * (1 - sx)))
def forward(self, x):
return self.F.apply(x)
# Mish https://github.com/digantamisra98/Mish --------------------------------------------------------------------------
class Mish(nn.Module):
@staticmethod
def forward(x):
return x * F.softplus(x).tanh()
class MemoryEfficientMish(nn.Module):
class F(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x)))
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_tensors[0]
sx = torch.sigmoid(x)
fx = F.softplus(x).tanh()
return grad_output * (fx + x * sx * (1 - fx * fx))
def forward(self, x):
return self.F.apply(x)
# FReLU https://arxiv.org/abs/2007.11824 -------------------------------------------------------------------------------
class FReLU(nn.Module):
def __init__(self, c1, k=3): # ch_in, kernel
super().__init__()
self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1)
self.bn = nn.BatchNorm2d(c1)
def forward(self, x):
return torch.max(x, self.bn(self.conv(x)))
utils/adabound.py
-236
View File
@@ -1,236 +0,0 @@
import math
import torch
from torch.optim.optimizer import Optimizer
class AdaBound(Optimizer):
"""Implements AdaBound algorithm.
It has been proposed in `Adaptive Gradient Methods with Dynamic Bound of Learning Rate`_.
Arguments:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float, optional): Adam learning rate (default: 1e-3)
betas (Tuple[float, float], optional): coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))
final_lr (float, optional): final (SGD) learning rate (default: 0.1)
gamma (float, optional): convergence speed of the bound functions (default: 1e-3)
eps (float, optional): term added to the denominator to improve
numerical stability (default: 1e-8)
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
amsbound (boolean, optional): whether to use the AMSBound variant of this algorithm
.. Adaptive Gradient Methods with Dynamic Bound of Learning Rate:
https://openreview.net/forum?id=Bkg3g2R9FX
"""
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), final_lr=0.1, gamma=1e-3,
eps=1e-8, weight_decay=0, amsbound=False):
if not 0.0 <= lr:
raise ValueError("Invalid learning rate: {}".format(lr))
if not 0.0 <= eps:
raise ValueError("Invalid epsilon value: {}".format(eps))
if not 0.0 <= betas[0] < 1.0:
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
if not 0.0 <= betas[1] < 1.0:
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
if not 0.0 <= final_lr:
raise ValueError("Invalid final learning rate: {}".format(final_lr))
if not 0.0 <= gamma < 1.0:
raise ValueError("Invalid gamma parameter: {}".format(gamma))
defaults = dict(lr=lr, betas=betas, final_lr=final_lr, gamma=gamma, eps=eps,
weight_decay=weight_decay, amsbound=amsbound)
super(AdaBound, self).__init__(params, defaults)
self.base_lrs = list(map(lambda group: group['lr'], self.param_groups))
def __setstate__(self, state):
super(AdaBound, self).__setstate__(state)
for group in self.param_groups:
group.setdefault('amsbound', False)
def step(self, closure=None):
"""Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
loss = closure()
for group, base_lr in zip(self.param_groups, self.base_lrs):
for p in group['params']:
if p.grad is None:
continue
grad = p.grad.data
if grad.is_sparse:
raise RuntimeError(
'Adam does not support sparse gradients, please consider SparseAdam instead')
amsbound = group['amsbound']
state = self.state[p]
# State initialization
if len(state) == 0:
state['step'] = 0
# Exponential moving average of gradient values
state['exp_avg'] = torch.zeros_like(p.data)
# Exponential moving average of squared gradient values
state['exp_avg_sq'] = torch.zeros_like(p.data)
if amsbound:
# Maintains max of all exp. moving avg. of sq. grad. values
state['max_exp_avg_sq'] = torch.zeros_like(p.data)
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
if amsbound:
max_exp_avg_sq = state['max_exp_avg_sq']
beta1, beta2 = group['betas']
state['step'] += 1
if group['weight_decay'] != 0:
grad = grad.add(group['weight_decay'], p.data)
# Decay the first and second moment running average coefficient
exp_avg.mul_(beta1).add_(1 - beta1, grad)
exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
if amsbound:
# Maintains the maximum of all 2nd moment running avg. till now
torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
# Use the max. for normalizing running avg. of gradient
denom = max_exp_avg_sq.sqrt().add_(group['eps'])
else:
denom = exp_avg_sq.sqrt().add_(group['eps'])
bias_correction1 = 1 - beta1 ** state['step']
bias_correction2 = 1 - beta2 ** state['step']
step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
# Applies bounds on actual learning rate
# lr_scheduler cannot affect final_lr, this is a workaround to apply lr decay
final_lr = group['final_lr'] * group['lr'] / base_lr
lower_bound = final_lr * (1 - 1 / (group['gamma'] * state['step'] + 1))
upper_bound = final_lr * (1 + 1 / (group['gamma'] * state['step']))
step_size = torch.full_like(denom, step_size)
step_size.div_(denom).clamp_(lower_bound, upper_bound).mul_(exp_avg)
p.data.add_(-step_size)
return loss
class AdaBoundW(Optimizer):
"""Implements AdaBound algorithm with Decoupled Weight Decay (arxiv.org/abs/1711.05101)
It has been proposed in `Adaptive Gradient Methods with Dynamic Bound of Learning Rate`_.
Arguments:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float, optional): Adam learning rate (default: 1e-3)
betas (Tuple[float, float], optional): coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))
final_lr (float, optional): final (SGD) learning rate (default: 0.1)
gamma (float, optional): convergence speed of the bound functions (default: 1e-3)
eps (float, optional): term added to the denominator to improve
numerical stability (default: 1e-8)
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
amsbound (boolean, optional): whether to use the AMSBound variant of this algorithm
.. Adaptive Gradient Methods with Dynamic Bound of Learning Rate:
https://openreview.net/forum?id=Bkg3g2R9FX
"""
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), final_lr=0.1, gamma=1e-3,
eps=1e-8, weight_decay=0, amsbound=False):
if not 0.0 <= lr:
raise ValueError("Invalid learning rate: {}".format(lr))
if not 0.0 <= eps:
raise ValueError("Invalid epsilon value: {}".format(eps))
if not 0.0 <= betas[0] < 1.0:
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
if not 0.0 <= betas[1] < 1.0:
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
if not 0.0 <= final_lr:
raise ValueError("Invalid final learning rate: {}".format(final_lr))
if not 0.0 <= gamma < 1.0:
raise ValueError("Invalid gamma parameter: {}".format(gamma))
defaults = dict(lr=lr, betas=betas, final_lr=final_lr, gamma=gamma, eps=eps,
weight_decay=weight_decay, amsbound=amsbound)
super(AdaBoundW, self).__init__(params, defaults)
self.base_lrs = list(map(lambda group: group['lr'], self.param_groups))
def __setstate__(self, state):
super(AdaBoundW, self).__setstate__(state)
for group in self.param_groups:
group.setdefault('amsbound', False)
def step(self, closure=None):
"""Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
loss = closure()
for group, base_lr in zip(self.param_groups, self.base_lrs):
for p in group['params']:
if p.grad is None:
continue
grad = p.grad.data
if grad.is_sparse:
raise RuntimeError(
'Adam does not support sparse gradients, please consider SparseAdam instead')
amsbound = group['amsbound']
state = self.state[p]
# State initialization
if len(state) == 0:
state['step'] = 0
# Exponential moving average of gradient values
state['exp_avg'] = torch.zeros_like(p.data)
# Exponential moving average of squared gradient values
state['exp_avg_sq'] = torch.zeros_like(p.data)
if amsbound:
# Maintains max of all exp. moving avg. of sq. grad. values
state['max_exp_avg_sq'] = torch.zeros_like(p.data)
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
if amsbound:
max_exp_avg_sq = state['max_exp_avg_sq']
beta1, beta2 = group['betas']
state['step'] += 1
# Decay the first and second moment running average coefficient
exp_avg.mul_(beta1).add_(1 - beta1, grad)
exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
if amsbound:
# Maintains the maximum of all 2nd moment running avg. till now
torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
# Use the max. for normalizing running avg. of gradient
denom = max_exp_avg_sq.sqrt().add_(group['eps'])
else:
denom = exp_avg_sq.sqrt().add_(group['eps'])
bias_correction1 = 1 - beta1 ** state['step']
bias_correction2 = 1 - beta2 ** state['step']
step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
# Applies bounds on actual learning rate
# lr_scheduler cannot affect final_lr, this is a workaround to apply lr decay
final_lr = group['final_lr'] * group['lr'] / base_lr
lower_bound = final_lr * (1 - 1 / (group['gamma'] * state['step'] + 1))
upper_bound = final_lr * (1 + 1 / (group['gamma'] * state['step']))
step_size = torch.full_like(denom, step_size)
step_size.div_(denom).clamp_(lower_bound, upper_bound).mul_(exp_avg)
if group['weight_decay'] != 0:
decayed_weights = torch.mul(p.data, group['weight_decay'])
p.data.add_(-step_size)
p.data.sub_(decayed_weights)
else:
p.data.add_(-step_size)
return loss
utils/autoanchor.py
+152
View File
@@ -0,0 +1,152 @@
# Auto-anchor utils
import numpy as np
import torch
import yaml
from scipy.cluster.vq import kmeans
from tqdm import tqdm
def check_anchor_order(m):
# Check anchor order against stride order for YOLOv3 Detect() module m, and correct if necessary
a = m.anchor_grid.prod(-1).view(-1) # anchor area
da = a[-1] - a[0] # delta a
ds = m.stride[-1] - m.stride[0] # delta s
if da.sign() != ds.sign(): # same order
print('Reversing anchor order')
m.anchors[:] = m.anchors.flip(0)
m.anchor_grid[:] = m.anchor_grid.flip(0)
def check_anchors(dataset, model, thr=4.0, imgsz=640):
# Check anchor fit to data, recompute if necessary
print('\nAnalyzing anchors... ', end='')
m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect()
shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True)
scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1)) # augment scale
wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float() # wh
def metric(k): # compute metric
r = wh[:, None] / k[None]
x = torch.min(r, 1. / r).min(2)[0] # ratio metric
best = x.max(1)[0] # best_x
aat = (x > 1. / thr).float().sum(1).mean() # anchors above threshold
bpr = (best > 1. / thr).float().mean() # best possible recall
return bpr, aat
bpr, aat = metric(m.anchor_grid.clone().cpu().view(-1, 2))
print('anchors/target = %.2f, Best Possible Recall (BPR) = %.4f' % (aat, bpr), end='')
if bpr < 0.98: # threshold to recompute
print('. Attempting to improve anchors, please wait...')
na = m.anchor_grid.numel() // 2 # number of anchors
new_anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
new_bpr = metric(new_anchors.reshape(-1, 2))[0]
if new_bpr > bpr: # replace anchors
new_anchors = torch.tensor(new_anchors, device=m.anchors.device).type_as(m.anchors)
m.anchor_grid[:] = new_anchors.clone().view_as(m.anchor_grid) # for inference
m.anchors[:] = new_anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1) # loss
check_anchor_order(m)
print('New anchors saved to model. Update model *.yaml to use these anchors in the future.')
else:
print('Original anchors better than new anchors. Proceeding with original anchors.')
print('') # newline
def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
""" Creates kmeans-evolved anchors from training dataset
Arguments:
path: path to dataset *.yaml, or a loaded dataset
n: number of anchors
img_size: image size used for training
thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0
gen: generations to evolve anchors using genetic algorithm
verbose: print all results
Return:
k: kmeans evolved anchors
Usage:
from utils.autoanchor import *; _ = kmean_anchors()
"""
thr = 1. / thr
def metric(k, wh): # compute metrics
r = wh[:, None] / k[None]
x = torch.min(r, 1. / r).min(2)[0] # ratio metric
# x = wh_iou(wh, torch.tensor(k)) # iou metric
return x, x.max(1)[0] # x, best_x
def anchor_fitness(k): # mutation fitness
_, best = metric(torch.tensor(k, dtype=torch.float32), wh)
return (best * (best > thr).float()).mean() # fitness
def print_results(k):
k = k[np.argsort(k.prod(1))] # sort small to large
x, best = metric(k, wh0)
bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n # best possible recall, anch > thr
print('thr=%.2f: %.4f best possible recall, %.2f anchors past thr' % (thr, bpr, aat))
print('n=%g, img_size=%s, metric_all=%.3f/%.3f-mean/best, past_thr=%.3f-mean: ' %
(n, img_size, x.mean(), best.mean(), x[x > thr].mean()), end='')
for i, x in enumerate(k):
print('%i,%i' % (round(x[0]), round(x[1])), end=', ' if i < len(k) - 1 else '\n') # use in *.cfg
return k
if isinstance(path, str): # *.yaml file
with open(path) as f:
data_dict = yaml.load(f, Loader=yaml.FullLoader) # model dict
from utils.datasets import LoadImagesAndLabels
dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
else:
dataset = path # dataset
# Get label wh
shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True)
wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)]) # wh
# Filter
i = (wh0 < 3.0).any(1).sum()
if i:
print('WARNING: Extremely small objects found. '
'%g of %g labels are < 3 pixels in width or height.' % (i, len(wh0)))
wh = wh0[(wh0 >= 2.0).any(1)] # filter > 2 pixels
# Kmeans calculation
print('Running kmeans for %g anchors on %g points...' % (n, len(wh)))
s = wh.std(0) # sigmas for whitening
k, dist = kmeans(wh / s, n, iter=30) # points, mean distance
k *= s
wh = torch.tensor(wh, dtype=torch.float32) # filtered
wh0 = torch.tensor(wh0, dtype=torch.float32) # unfiltered
k = print_results(k)
# Plot
# k, d = [None] * 20, [None] * 20
# for i in tqdm(range(1, 21)):
# k[i-1], d[i-1] = kmeans(wh / s, i) # points, mean distance
# fig, ax = plt.subplots(1, 2, figsize=(14, 7))
# ax = ax.ravel()
# ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
# fig, ax = plt.subplots(1, 2, figsize=(14, 7)) # plot wh
# ax[0].hist(wh[wh[:, 0]<100, 0],400)
# ax[1].hist(wh[wh[:, 1]<100, 1],400)
# fig.tight_layout()
# 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), desc='Evolving anchors with Genetic Algorithm') # progress bar
for _ in pbar:
v = np.ones(sh)
while (v == 1).all(): # mutate until a change occurs (prevent duplicates)
v = ((npr.random(sh) < mp) * npr.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
kg = (k.copy() * v).clip(min=2.0)
fg = anchor_fitness(kg)
if fg > f:
f, k = fg, kg.copy()
pbar.desc = 'Evolving anchors with Genetic Algorithm: fitness = %.4f' % f
if verbose:
print_results(k)
return print_results(k)
utils/datasets.py
+392 -304
View File
@@ -1,9 +1,14 @@
# Dataset utils and dataloaders
import glob
import logging
import math
import os
import random
import shutil
import time
from itertools import repeat
from multiprocessing.pool import ThreadPool
from pathlib import Path
from threading import Thread
@@ -14,11 +19,14 @@ from PIL import Image, ExifTags
from torch.utils.data import Dataset
from tqdm import tqdm
from utils.utils import xyxy2xywh, xywh2xyxy
from utils.general import xyxy2xywh, xywh2xyxy
from utils.torch_utils import torch_distributed_zero_first
# Parameters
help_url = 'https://github.com/ultralytics/yolov3/wiki/Train-Custom-Data'
img_formats = ['.bmp', '.jpg', '.jpeg', '.png', '.tif', '.tiff', '.dng']
vid_formats = ['.mov', '.avi', '.mp4', '.mpg', '.mpeg', '.m4v', '.wmv', '.mkv']
img_formats = ['bmp', 'jpg', 'jpeg', 'png', 'tif', 'tiff', 'dng'] # acceptable image suffixes
vid_formats = ['mov', 'avi', 'mp4', 'mpg', 'mpeg', 'm4v', 'wmv', 'mkv'] # acceptable video suffixes
logger = logging.getLogger(__name__)
# Get orientation exif tag
for orientation in ExifTags.TAGS.keys():
@@ -26,6 +34,11 @@ for orientation in ExifTags.TAGS.keys():
break
def get_hash(files):
# Returns a single hash value of a list of files
return sum(os.path.getsize(f) for f in files if os.path.isfile(f))
def exif_size(img):
# Returns exif-corrected PIL size
s = img.size # (width, height)
@@ -41,37 +54,104 @@ def exif_size(img):
return s
class LoadImages: # for inference
def __init__(self, path, img_size=416):
path = str(Path(path)) # os-agnostic
files = []
if os.path.isdir(path):
files = sorted(glob.glob(os.path.join(path, '*.*')))
elif os.path.isfile(path):
files = [path]
def create_dataloader(path, imgsz, batch_size, stride, opt, hyp=None, augment=False, cache=False, pad=0.0, rect=False,
rank=-1, world_size=1, workers=8, image_weights=False):
# Make sure only the first process in DDP process the dataset first, and the following others can use the cache
with torch_distributed_zero_first(rank):
dataset = LoadImagesAndLabels(path, imgsz, batch_size,
augment=augment, # augment images
hyp=hyp, # augmentation hyperparameters
rect=rect, # rectangular training
cache_images=cache,
single_cls=opt.single_cls,
stride=int(stride),
pad=pad,
rank=rank,
image_weights=image_weights)
images = [x for x in files if os.path.splitext(x)[-1].lower() in img_formats]
videos = [x for x in files if os.path.splitext(x)[-1].lower() in vid_formats]
nI, nV = len(images), len(videos)
batch_size = min(batch_size, len(dataset))
nw = min([os.cpu_count() // world_size, batch_size if batch_size > 1 else 0, workers]) # number of workers
sampler = torch.utils.data.distributed.DistributedSampler(dataset) if rank != -1 else None
loader = torch.utils.data.DataLoader if image_weights else InfiniteDataLoader
# Use torch.utils.data.DataLoader() if dataset.properties will update during training else InfiniteDataLoader()
dataloader = loader(dataset,
batch_size=batch_size,
num_workers=nw,
sampler=sampler,
pin_memory=True,
collate_fn=LoadImagesAndLabels.collate_fn)
return dataloader, dataset
class InfiniteDataLoader(torch.utils.data.dataloader.DataLoader):
""" Dataloader that reuses workers
Uses same syntax as vanilla DataLoader
"""
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
object.__setattr__(self, 'batch_sampler', _RepeatSampler(self.batch_sampler))
self.iterator = super().__iter__()
def __len__(self):
return len(self.batch_sampler.sampler)
def __iter__(self):
for i in range(len(self)):
yield next(self.iterator)
class _RepeatSampler(object):
""" Sampler that repeats forever
Args:
sampler (Sampler)
"""
def __init__(self, sampler):
self.sampler = sampler
def __iter__(self):
while True:
yield from iter(self.sampler)
class LoadImages: # for inference
def __init__(self, path, img_size=640):
p = str(Path(path)) # os-agnostic
p = os.path.abspath(p) # absolute path
if '*' in p:
files = sorted(glob.glob(p, recursive=True)) # glob
elif os.path.isdir(p):
files = sorted(glob.glob(os.path.join(p, '*.*'))) # dir
elif os.path.isfile(p):
files = [p] # files
else:
raise Exception('ERROR: %s does not exist' % p)
images = [x for x in files if x.split('.')[-1].lower() in img_formats]
videos = [x for x in files if x.split('.')[-1].lower() in vid_formats]
ni, nv = len(images), len(videos)
self.img_size = img_size
self.files = images + videos
self.nF = nI + nV # number of files
self.video_flag = [False] * nI + [True] * nV
self.nf = ni + nv # number of files
self.video_flag = [False] * ni + [True] * nv
self.mode = 'images'
if any(videos):
self.new_video(videos[0]) # new video
else:
self.cap = None
assert self.nF > 0, 'No images or videos found in %s. Supported formats are:\nimages: %s\nvideos: %s' % \
(path, img_formats, vid_formats)
assert self.nf > 0, 'No images or videos found in %s. Supported formats are:\nimages: %s\nvideos: %s' % \
(p, img_formats, vid_formats)
def __iter__(self):
self.count = 0
return self
def __next__(self):
if self.count == self.nF:
if self.count == self.nf:
raise StopIteration
path = self.files[self.count]
@@ -82,7 +162,7 @@ class LoadImages: # for inference
if not ret_val:
self.count += 1
self.cap.release()
if self.count == self.nF: # last video
if self.count == self.nf: # last video
raise StopIteration
else:
path = self.files[self.count]
@@ -90,14 +170,14 @@ class LoadImages: # for inference
ret_val, img0 = self.cap.read()
self.frame += 1
print('video %g/%g (%g/%g) %s: ' % (self.count + 1, self.nF, self.frame, self.nframes, path), end='')
print('video %g/%g (%g/%g) %s: ' % (self.count + 1, self.nf, self.frame, self.nframes, path), end='')
else:
# Read image
self.count += 1
img0 = cv2.imread(path) # BGR
assert img0 is not None, 'Image Not Found ' + path
print('image %g/%g %s: ' % (self.count, self.nF, path), end='')
print('image %g/%g %s: ' % (self.count, self.nf, path), end='')
# Padded resize
img = letterbox(img0, new_shape=self.img_size)[0]
@@ -106,7 +186,6 @@ class LoadImages: # for inference
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
# cv2.imwrite(path + '.letterbox.jpg', 255 * img.transpose((1, 2, 0))[:, :, ::-1]) # save letterbox image
return path, img, img0, self.cap
def new_video(self, path):
@@ -115,27 +194,19 @@ class LoadImages: # for inference
self.nframes = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT))
def __len__(self):
return self.nF # number of files
return self.nf # number of files
class LoadWebcam: # for inference
def __init__(self, pipe=0, img_size=416):
def __init__(self, pipe='0', img_size=640):
self.img_size = img_size
if pipe == '0':
pipe = 0 # local camera
if pipe.isnumeric():
pipe = eval(pipe) # local camera
# pipe = 'rtsp://192.168.1.64/1' # IP camera
# pipe = 'rtsp://username:password@192.168.1.64/1' # IP camera with login
# pipe = 'rtsp://170.93.143.139/rtplive/470011e600ef003a004ee33696235daa' # IP traffic camera
# pipe = 'http://wmccpinetop.axiscam.net/mjpg/video.mjpg' # IP golf camera
# https://answers.opencv.org/question/215996/changing-gstreamer-pipeline-to-opencv-in-pythonsolved/
# pipe = '"rtspsrc location="rtsp://username:password@192.168.1.64/1" latency=10 ! appsink' # GStreamer
# https://answers.opencv.org/question/200787/video-acceleration-gstremer-pipeline-in-videocapture/
# https://stackoverflow.com/questions/54095699/install-gstreamer-support-for-opencv-python-package # install help
# pipe = "rtspsrc location=rtsp://root:root@192.168.0.91:554/axis-media/media.amp?videocodec=h264&resolution=3840x2160 protocols=GST_RTSP_LOWER_TRANS_TCP ! rtph264depay ! queue ! vaapih264dec ! videoconvert ! appsink" # GStreamer
self.pipe = pipe
self.cap = cv2.VideoCapture(pipe) # video capture object
self.cap.set(cv2.CAP_PROP_BUFFERSIZE, 3) # set buffer size
@@ -184,7 +255,7 @@ class LoadWebcam: # for inference
class LoadStreams: # multiple IP or RTSP cameras
def __init__(self, sources='streams.txt', img_size=416):
def __init__(self, sources='streams.txt', img_size=640):
self.mode = 'images'
self.img_size = img_size
@@ -200,7 +271,7 @@ class LoadStreams: # multiple IP or RTSP cameras
for i, s in enumerate(sources):
# Start the thread to read frames from the video stream
print('%g/%g: %s... ' % (i + 1, n, s), end='')
cap = cv2.VideoCapture(0 if s == '0' else s)
cap = cv2.VideoCapture(eval(s) if s.isnumeric() else s)
assert cap.isOpened(), 'Failed to open %s' % s
w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
@@ -256,55 +327,77 @@ class LoadStreams: # multiple IP or RTSP cameras
return 0 # 1E12 frames = 32 streams at 30 FPS for 30 years
def img2label_paths(img_paths):
# Define label paths as a function of image paths
sa, sb = os.sep + 'images' + os.sep, os.sep + 'labels' + os.sep # /images/, /labels/ substrings
return [x.replace(sa, sb, 1).replace('.' + x.split('.')[-1], '.txt') for x in img_paths]
class LoadImagesAndLabels(Dataset): # for training/testing
def __init__(self, path, img_size=416, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False,
cache_images=False, single_cls=False, pad=0.0):
try:
path = str(Path(path)) # os-agnostic
parent = str(Path(path).parent) + os.sep
if os.path.isfile(path): # file
with open(path, 'r') as f:
f = f.read().splitlines()
f = [x.replace('./', parent) if x.startswith('./') else x for x in f] # local to global path
elif os.path.isdir(path): # folder
f = glob.iglob(path + os.sep + '*.*')
else:
raise Exception('%s does not exist' % path)
self.img_files = [x.replace('/', os.sep) for x in f if os.path.splitext(x)[-1].lower() in img_formats]
except:
raise Exception('Error loading data from %s. See %s' % (path, help_url))
n = len(self.img_files)
assert n > 0, 'No images found in %s. See %s' % (path, help_url)
bi = np.floor(np.arange(n) / batch_size).astype(np.int) # batch index
nb = bi[-1] + 1 # number of batches
self.n = n # number of images
self.batch = bi # batch index of image
def __init__(self, path, img_size=640, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False,
cache_images=False, single_cls=False, stride=32, pad=0.0, rank=-1):
self.img_size = img_size
self.augment = augment
self.hyp = hyp
self.image_weights = image_weights
self.rect = False if image_weights else rect
self.mosaic = self.augment and not self.rect # load 4 images at a time into a mosaic (only during training)
self.mosaic_border = [-img_size // 2, -img_size // 2]
self.stride = stride
# Define labels
self.label_files = [x.replace('images', 'labels').replace(os.path.splitext(x)[-1], '.txt')
for x in self.img_files]
# Read image shapes (wh)
sp = path.replace('.txt', '') + '.shapes' # shapefile path
try:
with open(sp, 'r') as f: # read existing shapefile
s = [x.split() for x in f.read().splitlines()]
assert len(s) == n, 'Shapefile out of sync'
except:
s = [exif_size(Image.open(f)) for f in tqdm(self.img_files, desc='Reading image shapes')]
np.savetxt(sp, s, fmt='%g') # overwrites existing (if any)
f = [] # image files
for p in path if isinstance(path, list) else [path]:
p = Path(p) # os-agnostic
if p.is_dir(): # dir
f += glob.glob(str(p / '**' / '*.*'), recursive=True)
elif p.is_file(): # file
with open(p, 'r') as t:
t = t.read().splitlines()
parent = str(p.parent) + os.sep
f += [x.replace('./', parent) if x.startswith('./') else x for x in t] # local to global path
else:
raise Exception('%s does not exist' % p)
self.img_files = sorted([x.replace('/', os.sep) for x in f if x.split('.')[-1].lower() in img_formats])
assert self.img_files, 'No images found'
except Exception as e:
raise Exception('Error loading data from %s: %s\nSee %s' % (path, e, help_url))
self.shapes = np.array(s, dtype=np.float64)
# Check cache
self.label_files = img2label_paths(self.img_files) # labels
cache_path = Path(self.label_files[0]).parent.with_suffix('.cache') # cached labels
if cache_path.is_file():
cache = torch.load(cache_path) # load
if cache['hash'] != get_hash(self.label_files + self.img_files) or 'results' not in cache: # changed
cache = self.cache_labels(cache_path) # re-cache
else:
cache = self.cache_labels(cache_path) # cache
# Rectangular Training https://github.com/ultralytics/yolov3/issues/232
# Display cache
[nf, nm, ne, nc, n] = cache.pop('results') # found, missing, empty, corrupted, total
desc = f"Scanning '{cache_path}' for images and labels... {nf} found, {nm} missing, {ne} empty, {nc} corrupted"
tqdm(None, desc=desc, total=n, initial=n)
assert nf > 0 or not augment, f'No labels found in {cache_path}. Can not train without labels. See {help_url}'
# Read cache
cache.pop('hash') # remove hash
labels, shapes = zip(*cache.values())
self.labels = list(labels)
self.shapes = np.array(shapes, dtype=np.float64)
self.img_files = list(cache.keys()) # update
self.label_files = img2label_paths(cache.keys()) # update
if single_cls:
for x in self.labels:
x[:, 0] = 0
n = len(shapes) # number of images
bi = np.floor(np.arange(n) / batch_size).astype(np.int) # batch index
nb = bi[-1] + 1 # number of batches
self.batch = bi # batch index of image
self.n = n
self.indices = range(n)
# Rectangular Training
if self.rect:
# Sort by aspect ratio
s = self.shapes # wh
@@ -312,6 +405,7 @@ class LoadImagesAndLabels(Dataset): # for training/testing
irect = ar.argsort()
self.img_files = [self.img_files[i] for i in irect]
self.label_files = [self.label_files[i] for i in irect]
self.labels = [self.labels[i] for i in irect]
self.shapes = s[irect] # wh
ar = ar[irect]
@@ -325,107 +419,65 @@ class LoadImagesAndLabels(Dataset): # for training/testing
elif mini > 1:
shapes[i] = [1, 1 / mini]
self.batch_shapes = np.ceil(np.array(shapes) * img_size / 32. + pad).astype(np.int) * 32
# Cache labels
self.imgs = [None] * n
self.labels = [np.zeros((0, 5), dtype=np.float32)] * n
create_datasubset, extract_bounding_boxes, labels_loaded = False, False, False
nm, nf, ne, ns, nd = 0, 0, 0, 0, 0 # number missing, found, empty, datasubset, duplicate
np_labels_path = str(Path(self.label_files[0]).parent) + '.npy' # saved labels in *.npy file
if os.path.isfile(np_labels_path):
s = np_labels_path # print string
x = np.load(np_labels_path, allow_pickle=True)
if len(x) == n:
self.labels = x
labels_loaded = True
else:
s = path.replace('images', 'labels')
pbar = tqdm(self.label_files)
for i, file in enumerate(pbar):
if labels_loaded:
l = self.labels[i]
# np.savetxt(file, l, '%g') # save *.txt from *.npy file
else:
try:
with open(file, 'r') as f:
l = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32)
except:
nm += 1 # print('missing labels for image %s' % self.img_files[i]) # file missing
continue
if l.shape[0]:
assert l.shape[1] == 5, '> 5 label columns: %s' % file
assert (l >= 0).all(), 'negative labels: %s' % file
assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels: %s' % file
if np.unique(l, axis=0).shape[0] < l.shape[0]: # duplicate rows
nd += 1 # print('WARNING: duplicate rows in %s' % self.label_files[i]) # duplicate rows
if single_cls:
l[:, 0] = 0 # force dataset into single-class mode
self.labels[i] = l
nf += 1 # file found
# Create subdataset (a smaller dataset)
if create_datasubset and ns < 1E4:
if ns == 0:
create_folder(path='./datasubset')
os.makedirs('./datasubset/images')
exclude_classes = 43
if exclude_classes not in l[:, 0]:
ns += 1
# shutil.copy(src=self.img_files[i], dst='./datasubset/images/') # copy image
with open('./datasubset/images.txt', 'a') as f:
f.write(self.img_files[i] + '\n')
# Extract object detection boxes for a second stage classifier
if extract_bounding_boxes:
p = Path(self.img_files[i])
img = cv2.imread(str(p))
h, w = img.shape[:2]
for j, x in enumerate(l):
f = '%s%sclassifier%s%g_%g_%s' % (p.parent.parent, os.sep, os.sep, x[0], j, p.name)
if not os.path.exists(Path(f).parent):
os.makedirs(Path(f).parent) # make new output folder
b = x[1:] * [w, h, w, h] # box
b[2:] = b[2:].max() # rectangle to square
b[2:] = b[2:] * 1.3 + 30 # pad
b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int)
b[[0, 2]] = np.clip(b[[0, 2]], 0, w) # clip boxes outside of image
b[[1, 3]] = np.clip(b[[1, 3]], 0, h)
assert cv2.imwrite(f, img[b[1]:b[3], b[0]:b[2]]), 'Failure extracting classifier boxes'
else:
ne += 1 # print('empty labels for image %s' % self.img_files[i]) # file empty
# os.system("rm '%s' '%s'" % (self.img_files[i], self.label_files[i])) # remove
pbar.desc = 'Caching labels %s (%g found, %g missing, %g empty, %g duplicate, for %g images)' % (
s, nf, nm, ne, nd, n)
assert nf > 0 or n == 20288, 'No labels found in %s. See %s' % (os.path.dirname(file) + os.sep, help_url)
if not labels_loaded and n > 1000:
print('Saving labels to %s for faster future loading' % np_labels_path)
np.save(np_labels_path, self.labels) # save for next time
self.batch_shapes = np.ceil(np.array(shapes) * img_size / stride + pad).astype(np.int) * stride
# Cache images into memory for faster training (WARNING: large datasets may exceed system RAM)
if cache_images: # if training
self.imgs = [None] * n
if cache_images:
gb = 0 # Gigabytes of cached images
pbar = tqdm(range(len(self.img_files)), desc='Caching images')
self.img_hw0, self.img_hw = [None] * n, [None] * n
for i in pbar: # max 10k images
self.imgs[i], self.img_hw0[i], self.img_hw[i] = load_image(self, i) # img, hw_original, hw_resized
results = ThreadPool(8).imap(lambda x: load_image(*x), zip(repeat(self), range(n))) # 8 threads
pbar = tqdm(enumerate(results), total=n)
for i, x in pbar:
self.imgs[i], self.img_hw0[i], self.img_hw[i] = x # img, hw_original, hw_resized = load_image(self, i)
gb += self.imgs[i].nbytes
pbar.desc = 'Caching images (%.1fGB)' % (gb / 1E9)
# Detect corrupted images https://medium.com/joelthchao/programmatically-detect-corrupted-image-8c1b2006c3d3
detect_corrupted_images = False
if detect_corrupted_images:
from skimage import io # conda install -c conda-forge scikit-image
for file in tqdm(self.img_files, desc='Detecting corrupted images'):
try:
_ = io.imread(file)
except:
print('Corrupted image detected: %s' % file)
def cache_labels(self, path=Path('./labels.cache')):
# Cache dataset labels, check images and read shapes
x = {} # dict
nm, nf, ne, nc = 0, 0, 0, 0 # number missing, found, empty, duplicate
pbar = tqdm(zip(self.img_files, self.label_files), desc='Scanning images', total=len(self.img_files))
for i, (im_file, lb_file) in enumerate(pbar):
try:
# verify images
im = Image.open(im_file)
im.verify() # PIL verify
shape = exif_size(im) # image size
assert (shape[0] > 9) & (shape[1] > 9), 'image size <10 pixels'
# verify labels
if os.path.isfile(lb_file):
nf += 1 # label found
with open(lb_file, 'r') as f:
l = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32) # labels
if len(l):
assert l.shape[1] == 5, 'labels require 5 columns each'
assert (l >= 0).all(), 'negative labels'
assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels'
assert np.unique(l, axis=0).shape[0] == l.shape[0], 'duplicate labels'
else:
ne += 1 # label empty
l = np.zeros((0, 5), dtype=np.float32)
else:
nm += 1 # label missing
l = np.zeros((0, 5), dtype=np.float32)
x[im_file] = [l, shape]
except Exception as e:
nc += 1
print('WARNING: Ignoring corrupted image and/or label %s: %s' % (im_file, e))
pbar.desc = f"Scanning '{path.parent / path.stem}' for images and labels... " \
f"{nf} found, {nm} missing, {ne} empty, {nc} corrupted"
if nf == 0:
print(f'WARNING: No labels found in {path}. See {help_url}')
x['hash'] = get_hash(self.label_files + self.img_files)
x['results'] = [nf, nm, ne, nc, i + 1]
torch.save(x, path) # save for next time
logging.info(f"New cache created: {path}")
return x
def __len__(self):
return len(self.img_files)
@@ -437,15 +489,22 @@ class LoadImagesAndLabels(Dataset): # for training/testing
# return self
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
index = self.indices[index] # linear, shuffled, or image_weights
hyp = self.hyp
if self.mosaic:
mosaic = self.mosaic and random.random() < hyp['mosaic']
if mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
shapes = None
# MixUp https://arxiv.org/pdf/1710.09412.pdf
if random.random() < hyp['mixup']:
img2, labels2 = load_mosaic(self, random.randint(0, self.n - 1))
r = np.random.beta(8.0, 8.0) # mixup ratio, alpha=beta=8.0
img = (img * r + img2 * (1 - r)).astype(np.uint8)
labels = np.concatenate((labels, labels2), 0)
else:
# Load image
img, (h0, w0), (h, w) = load_image(self, index)
@@ -468,12 +527,13 @@ class LoadImagesAndLabels(Dataset): # for training/testing
if self.augment:
# Augment imagespace
if not self.mosaic:
img, labels = random_affine(img, labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
if not mosaic:
img, labels = random_perspective(img, labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'],
perspective=hyp['perspective'])
# Augment colorspace
augment_hsv(img, hgain=hyp['hsv_h'], sgain=hyp['hsv_s'], vgain=hyp['hsv_v'])
@@ -484,28 +544,23 @@ class LoadImagesAndLabels(Dataset): # for training/testing
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5]) # convert xyxy to xywh
labels[:, [2, 4]] /= img.shape[0] # normalized height 0-1
labels[:, [1, 3]] /= img.shape[1] # normalized width 0-1
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
# flip up-down
if random.random() < hyp['flipud']:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
# flip left-right
if random.random() < hyp['fliplr']:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
@@ -524,6 +579,7 @@ class LoadImagesAndLabels(Dataset): # for training/testing
return torch.stack(img, 0), torch.cat(label, 0), path, shapes
# Ancillary functions --------------------------------------------------------------------------------------------------
def load_image(self, index):
# loads 1 image from dataset, returns img, original hw, resized hw
img = self.imgs[index]
@@ -565,8 +621,8 @@ def load_mosaic(self, index):
labels4 = []
s = self.img_size
xc, yc = [int(random.uniform(s * 0.5, s * 1.5)) for _ in range(2)] # mosaic center x, y
indices = [index] + [random.randint(0, len(self.labels) - 1) for _ in range(3)] # 3 additional image indices
yc, xc = [int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border] # mosaic center x, y
indices = [index] + [self.indices[random.randint(0, self.n - 1)] for _ in range(3)] # 3 additional image indices
for i, index in enumerate(indices):
# Load image
img, _, (h, w) = load_image(self, index)
@@ -581,7 +637,7 @@ def load_mosaic(self, index):
x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
elif i == 2: # bottom left
x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, max(xc, w), min(y2a - y1a, h)
x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
elif i == 3: # bottom right
x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
@@ -603,22 +659,39 @@ def load_mosaic(self, index):
# Concat/clip labels
if len(labels4):
labels4 = np.concatenate(labels4, 0)
# np.clip(labels4[:, 1:] - s / 2, 0, s, out=labels4[:, 1:]) # use with center crop
np.clip(labels4[:, 1:], 0, 2 * s, out=labels4[:, 1:]) # use with random_affine
np.clip(labels4[:, 1:], 0, 2 * s, out=labels4[:, 1:]) # use with random_perspective
# img4, labels4 = replicate(img4, labels4) # replicate
# Augment
# img4 = img4[s // 2: int(s * 1.5), s // 2:int(s * 1.5)] # center crop (WARNING, requires box pruning)
img4, labels4 = random_affine(img4, labels4,
degrees=self.hyp['degrees'],
translate=self.hyp['translate'],
scale=self.hyp['scale'],
shear=self.hyp['shear'],
border=-s // 2) # border to remove
img4, labels4 = random_perspective(img4, labels4,
degrees=self.hyp['degrees'],
translate=self.hyp['translate'],
scale=self.hyp['scale'],
shear=self.hyp['shear'],
perspective=self.hyp['perspective'],
border=self.mosaic_border) # border to remove
return img4, labels4
def letterbox(img, new_shape=(416, 416), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True):
def replicate(img, labels):
# Replicate labels
h, w = img.shape[:2]
boxes = labels[:, 1:].astype(int)
x1, y1, x2, y2 = boxes.T
s = ((x2 - x1) + (y2 - y1)) / 2 # side length (pixels)
for i in s.argsort()[:round(s.size * 0.5)]: # smallest indices
x1b, y1b, x2b, y2b = boxes[i]
bh, bw = y2b - y1b, x2b - x1b
yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw)) # offset x, y
x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
img[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax]
labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
return img, labels
def letterbox(img, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True):
# Resize image to a 32-pixel-multiple rectangle https://github.com/ultralytics/yolov3/issues/232
shape = img.shape[:2] # current shape [height, width]
if isinstance(new_shape, int):
@@ -637,8 +710,8 @@ def letterbox(img, new_shape=(416, 416), color=(114, 114, 114), auto=True, scale
dw, dh = np.mod(dw, 32), np.mod(dh, 32) # wh padding
elif scaleFill: # stretch
dw, dh = 0.0, 0.0
new_unpad = new_shape
ratio = new_shape[0] / shape[1], new_shape[1] / shape[0] # width, height ratios
new_unpad = (new_shape[1], new_shape[0])
ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # width, height ratios
dw /= 2 # divide padding into 2 sides
dh /= 2
@@ -651,13 +724,22 @@ def letterbox(img, new_shape=(416, 416), color=(114, 114, 114), auto=True, scale
return img, ratio, (dw, dh)
def random_affine(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10, border=0):
def random_perspective(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10, perspective=0.0, border=(0, 0)):
# torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(.1, .1), scale=(.9, 1.1), shear=(-10, 10))
# https://medium.com/uruvideo/dataset-augmentation-with-random-homographies-a8f4b44830d4
# targets = [cls, xyxy]
height = img.shape[0] + border * 2
width = img.shape[1] + border * 2
height = img.shape[0] + border[0] * 2 # shape(h,w,c)
width = img.shape[1] + border[1] * 2
# Center
C = np.eye(3)
C[0, 2] = -img.shape[1] / 2 # x translation (pixels)
C[1, 2] = -img.shape[0] / 2 # y translation (pixels)
# Perspective
P = np.eye(3)
P[2, 0] = random.uniform(-perspective, perspective) # x perspective (about y)
P[2, 1] = random.uniform(-perspective, perspective) # y perspective (about x)
# Rotation and Scale
R = np.eye(3)
@@ -665,22 +747,31 @@ def random_affine(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10,
# a += random.choice([-180, -90, 0, 90]) # add 90deg rotations to small rotations
s = random.uniform(1 - scale, 1 + scale)
# s = 2 ** random.uniform(-scale, scale)
R[:2] = cv2.getRotationMatrix2D(angle=a, center=(img.shape[1] / 2, img.shape[0] / 2), scale=s)
# Translation
T = np.eye(3)
T[0, 2] = random.uniform(-translate, translate) * img.shape[0] + border # x translation (pixels)
T[1, 2] = random.uniform(-translate, translate) * img.shape[1] + border # y translation (pixels)
R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
# Shear
S = np.eye(3)
S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # x shear (deg)
S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # y shear (deg)
# Translation
T = np.eye(3)
T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width # x translation (pixels)
T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height # y translation (pixels)
# Combined rotation matrix
M = S @ T @ R # ORDER IS IMPORTANT HERE!!
if (border != 0) or (M != np.eye(3)).any(): # image changed
img = cv2.warpAffine(img, M[:2], dsize=(width, height), flags=cv2.INTER_LINEAR, borderValue=(114, 114, 114))
M = T @ S @ R @ P @ C # order of operations (right to left) is IMPORTANT
if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any(): # image changed
if perspective:
img = cv2.warpPerspective(img, M, dsize=(width, height), borderValue=(114, 114, 114))
else: # affine
img = cv2.warpAffine(img, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
# Visualize
# import matplotlib.pyplot as plt
# ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
# ax[0].imshow(img[:, :, ::-1]) # base
# ax[1].imshow(img2[:, :, ::-1]) # warped
# Transform label coordinates
n = len(targets)
@@ -688,7 +779,11 @@ def random_affine(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10,
# warp points
xy = np.ones((n * 4, 3))
xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1
xy = (xy @ M.T)[:, :2].reshape(n, 8)
xy = xy @ M.T # transform
if perspective:
xy = (xy[:, :2] / xy[:, 2:3]).reshape(n, 8) # rescale
else: # affine
xy = xy[:, :2].reshape(n, 8)
# create new boxes
x = xy[:, [0, 2, 4, 6]]
@@ -704,26 +799,28 @@ def random_affine(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10,
# h = (xy[:, 3] - xy[:, 1]) * reduction
# xy = np.concatenate((x - w / 2, y - h / 2, x + w / 2, y + h / 2)).reshape(4, n).T
# reject warped points outside of image
# clip boxes
xy[:, [0, 2]] = xy[:, [0, 2]].clip(0, width)
xy[:, [1, 3]] = xy[:, [1, 3]].clip(0, height)
w = xy[:, 2] - xy[:, 0]
h = xy[:, 3] - xy[:, 1]
area = w * h
area0 = (targets[:, 3] - targets[:, 1]) * (targets[:, 4] - targets[:, 2])
ar = np.maximum(w / (h + 1e-16), h / (w + 1e-16)) # aspect ratio
i = (w > 4) & (h > 4) & (area / (area0 * s + 1e-16) > 0.2) & (ar < 10)
# filter candidates
i = box_candidates(box1=targets[:, 1:5].T * s, box2=xy.T)
targets = targets[i]
targets[:, 1:5] = xy[i]
return img, targets
def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.1): # 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 + 1e-16), h2 / (w2 + 1e-16)) # aspect ratio
return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + 1e-16) > area_thr) & (ar < ar_thr) # candidates
def cutout(image, labels):
# https://arxiv.org/abs/1708.04552
# https://github.com/hysts/pytorch_cutout/blob/master/dataloader.py
# https://towardsdatascience.com/when-conventional-wisdom-fails-revisiting-data-augmentation-for-self-driving-cars-4831998c5509
# Applies image cutout augmentation https://arxiv.org/abs/1708.04552
h, w = image.shape[:2]
def bbox_ioa(box1, box2):
@@ -768,78 +865,69 @@ def cutout(image, labels):
return labels
def reduce_img_size(path='../data/sm4/images', img_size=1024): # from utils.datasets import *; reduce_img_size()
# creates a new ./images_reduced folder with reduced size images of maximum size img_size
path_new = path + '_reduced' # reduced images path
create_folder(path_new)
for f in tqdm(glob.glob('%s/*.*' % path)):
try:
img = cv2.imread(f)
h, w = img.shape[:2]
r = img_size / max(h, w) # size ratio
if r < 1.0:
img = cv2.resize(img, (int(w * r), int(h * r)), interpolation=cv2.INTER_AREA) # _LINEAR fastest
fnew = f.replace(path, path_new) # .replace(Path(f).suffix, '.jpg')
cv2.imwrite(fnew, img)
except:
print('WARNING: image failure %s' % f)
def convert_images2bmp(): # from utils.datasets import *; convert_images2bmp()
# Save images
formats = [x.lower() for x in img_formats] + [x.upper() for x in img_formats]
# for path in ['../coco/images/val2014', '../coco/images/train2014']:
for path in ['../data/sm4/images', '../data/sm4/background']:
create_folder(path + 'bmp')
for ext in formats: # ['.bmp', '.jpg', '.jpeg', '.png', '.tif', '.dng']
for f in tqdm(glob.glob('%s/*%s' % (path, ext)), desc='Converting %s' % ext):
cv2.imwrite(f.replace(ext.lower(), '.bmp').replace(path, path + 'bmp'), cv2.imread(f))
# Save labels
# for path in ['../coco/trainvalno5k.txt', '../coco/5k.txt']:
for file in ['../data/sm4/out_train.txt', '../data/sm4/out_test.txt']:
with open(file, 'r') as f:
lines = f.read()
# lines = f.read().replace('2014/', '2014bmp/') # coco
lines = lines.replace('/images', '/imagesbmp')
lines = lines.replace('/background', '/backgroundbmp')
for ext in formats:
lines = lines.replace(ext, '.bmp')
with open(file.replace('.txt', 'bmp.txt'), 'w') as f:
f.write(lines)
def recursive_dataset2bmp(dataset='../data/sm4_bmp'): # from utils.datasets import *; recursive_dataset2bmp()
# Converts dataset to bmp (for faster training)
formats = [x.lower() for x in img_formats] + [x.upper() for x in img_formats]
for a, b, files in os.walk(dataset):
for file in tqdm(files, desc=a):
p = a + '/' + file
s = Path(file).suffix
if s == '.txt': # replace text
with open(p, 'r') as f:
lines = f.read()
for f in formats:
lines = lines.replace(f, '.bmp')
with open(p, 'w') as f:
f.write(lines)
elif s in formats: # replace image
cv2.imwrite(p.replace(s, '.bmp'), cv2.imread(p))
if s != '.bmp':
os.system("rm '%s'" % p)
def imagelist2folder(path='data/coco_64img.txt'): # from utils.datasets import *; imagelist2folder()
# Copies all the images in a text file (list of images) into a folder
create_folder(path[:-4])
with open(path, 'r') as f:
for line in f.read().splitlines():
os.system('cp "%s" %s' % (line, path[:-4]))
print(line)
def create_folder(path='./new_folder'):
def create_folder(path='./new'):
# Create folder
if os.path.exists(path):
shutil.rmtree(path) # delete output folder
os.makedirs(path) # make new output folder
def flatten_recursive(path='../coco128'):
# Flatten a recursive directory by bringing all files to top level
new_path = Path(path + '_flat')
create_folder(new_path)
for file in tqdm(glob.glob(str(Path(path)) + '/**/*.*', recursive=True)):
shutil.copyfile(file, new_path / Path(file).name)
def extract_boxes(path='../coco128/'): # from utils.datasets import *; extract_boxes('../coco128')
# Convert detection dataset into classification dataset, with one directory per class
path = Path(path) # images dir
shutil.rmtree(path / 'classifier') if (path / 'classifier').is_dir() else None # remove existing
files = list(path.rglob('*.*'))
n = len(files) # number of files
for im_file in tqdm(files, total=n):
if im_file.suffix[1:] in img_formats:
# image
im = cv2.imread(str(im_file))[..., ::-1] # BGR to RGB
h, w = im.shape[:2]
# labels
lb_file = Path(img2label_paths([str(im_file)])[0])
if Path(lb_file).exists():
with open(lb_file, 'r') as f:
lb = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32) # labels
for j, x in enumerate(lb):
c = int(x[0]) # class
f = (path / 'classifier') / f'{c}' / f'{path.stem}_{im_file.stem}_{j}.jpg' # new filename
if not f.parent.is_dir():
f.parent.mkdir(parents=True)
b = x[1:] * [w, h, w, h] # box
# b[2:] = b[2:].max() # rectangle to square
b[2:] = b[2:] * 1.2 + 3 # pad
b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int)
b[[0, 2]] = np.clip(b[[0, 2]], 0, w) # clip boxes outside of image
b[[1, 3]] = np.clip(b[[1, 3]], 0, h)
assert cv2.imwrite(str(f), im[b[1]:b[3], b[0]:b[2]]), f'box failure in {f}'
def autosplit(path='../coco128', weights=(0.9, 0.1, 0.0)): # from utils.datasets import *; autosplit('../coco128')
""" Autosplit a dataset into train/val/test splits and save path/autosplit_*.txt files
# Arguments
path: Path to images directory
weights: Train, val, test weights (list)
"""
path = Path(path) # images dir
files = list(path.rglob('*.*'))
n = len(files) # number of files
indices = random.choices([0, 1, 2], weights=weights, k=n) # assign each image to a split
txt = ['autosplit_train.txt', 'autosplit_val.txt', 'autosplit_test.txt'] # 3 txt files
[(path / x).unlink() for x in txt if (path / x).exists()] # remove existing
for i, img in tqdm(zip(indices, files), total=n):
if img.suffix[1:] in img_formats:
with open(path / txt[i], 'a') as f:
f.write(str(img) + '\n') # add image to txt file
utils/evolve.sh
-18
View File
@@ -1,18 +0,0 @@
#!/bin/bash
#for i in 0 1 2 3
#do
# t=ultralytics/yolov3:v139 && sudo docker pull $t && sudo nvidia-docker run -d --ipc=host -v "$(pwd)"/coco:/usr/src/coco $t utils/evolve.sh $i
# sleep 30
#done
while true; do
# python3 train.py --data ../data/sm4/out.data --img-size 320 --epochs 100 --batch 64 --accum 1 --weights yolov3-tiny.conv.15 --multi --bucket ult/wer --evolve --cache --device $1 --cfg yolov3-tiny3-1cls.cfg --single --adam
# python3 train.py --data ../out/data.data --img-size 608 --epochs 10 --batch 8 --accum 8 --weights ultralytics68.pt --multi --bucket ult/athena --evolve --device $1 --cfg yolov3-spp-1cls.cfg
python3 train.py --data coco2014.data --img-size 512 608 --epochs 27 --batch 8 --accum 8 --evolve --weights '' --bucket ult/coco/sppa_512 --device $1 --cfg yolov3-sppa.cfg --multi
done
# coco epoch times --img-size 416 608 --epochs 27 --batch 16 --accum 4
# 36:34 2080ti
# 21:58 V100
# 63:00 T4
utils/gcp.sh
-39
View File
@@ -1,39 +0,0 @@
#!/usr/bin/env bash
# New VM
rm -rf sample_data yolov3
git clone https://github.com/ultralytics/yolov3
# git clone -b test --depth 1 https://github.com/ultralytics/yolov3 test # branch
# sudo apt-get install zip
#git clone https://github.com/NVIDIA/apex && cd apex && pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" . --user && cd .. && rm -rf apex
sudo conda install -yc conda-forge scikit-image pycocotools
# python3 -c "from yolov3.utils.google_utils import gdrive_download; gdrive_download('193Zp_ye-3qXMonR1nZj3YyxMtQkMy50k','coco2014.zip')"
python3 -c "from yolov3.utils.google_utils import gdrive_download; gdrive_download('1WQT6SOktSe8Uw6r10-2JhbEhMY5DJaph','coco2017.zip')"
python3 -c "from yolov3.utils.google_utils import gdrive_download; gdrive_download('1C3HewOG9akA3y456SZLBJZfNDPkBwAto','knife.zip')"
python3 -c "from yolov3.utils.google_utils import gdrive_download; gdrive_download('13g3LqdpkNE8sPosVJT6KFXlfoMypzRP4','sm4.zip')"
sudo shutdown
# Mount local SSD
lsblk
sudo mkfs.ext4 -F /dev/nvme0n1
sudo mkdir -p /mnt/disks/nvme0n1
sudo mount /dev/nvme0n1 /mnt/disks/nvme0n1
sudo chmod a+w /mnt/disks/nvme0n1
cp -r coco /mnt/disks/nvme0n1
# Kill All
t=ultralytics/yolov3:v1
docker kill $(docker ps -a -q --filter ancestor=$t)
# Evolve coco
sudo -s
t=ultralytics/yolov3:evolve
# docker kill $(docker ps -a -q --filter ancestor=$t)
for i in 0 1 6 7; do
docker pull $t && docker run --gpus all -d --ipc=host -v "$(pwd)"/coco:/usr/src/coco $t bash utils/evolve.sh $i
sleep 30
done
#COCO training
n=131 && t=ultralytics/coco:v131 && sudo docker pull $t && sudo docker run -it --gpus all --ipc=host -v "$(pwd)"/coco:/usr/src/coco $t python3 train.py --data coco2014.data --img-size 320 640 --epochs 300 --batch 16 --weights '' --device 0 --cfg yolov3-spp.cfg --bucket ult/coco --name $n && sudo shutdown
n=132 && t=ultralytics/coco:v131 && sudo docker pull $t && sudo docker run -it --gpus all --ipc=host -v "$(pwd)"/coco:/usr/src/coco $t python3 train.py --data coco2014.data --img-size 320 640 --epochs 300 --batch 64 --weights '' --device 0 --cfg yolov3-tiny.cfg --bucket ult/coco --name $n && sudo shutdown
utils/general.py
Executable
+445
View File
@@ -0,0 +1,445 @@
# General utils
import glob
import logging
import math
import os
import platform
import random
import re
import subprocess
import time
from pathlib import Path
import cv2
import matplotlib
import numpy as np
import torch
import torchvision
import yaml
from utils.google_utils import gsutil_getsize
from utils.metrics import fitness
from utils.torch_utils import init_torch_seeds
# Set printoptions
torch.set_printoptions(linewidth=320, precision=5, profile='long')
np.set_printoptions(linewidth=320, formatter={'float_kind': '{:11.5g}'.format}) # format short g, %precision=5
matplotlib.rc('font', **{'size': 11})
# Prevent OpenCV from multithreading (to use PyTorch DataLoader)
cv2.setNumThreads(0)
def set_logging(rank=-1):
logging.basicConfig(
format="%(message)s",
level=logging.INFO if rank in [-1, 0] else logging.WARN)
def init_seeds(seed=0):
random.seed(seed)
np.random.seed(seed)
init_torch_seeds(seed)
def get_latest_run(search_dir='.'):
# Return path to most recent 'last.pt' in /runs (i.e. to --resume from)
last_list = glob.glob(f'{search_dir}/**/last*.pt', recursive=True)
return max(last_list, key=os.path.getctime) if last_list else ''
def check_git_status():
# Suggest 'git pull' if repo is out of date
if platform.system() in ['Linux', 'Darwin'] and not os.path.isfile('/.dockerenv'):
s = subprocess.check_output('if [ -d .git ]; then git fetch && git status -uno; fi', shell=True).decode('utf-8')
if 'Your branch is behind' in s:
print(s[s.find('Your branch is behind'):s.find('\n\n')] + '\n')
def check_img_size(img_size, s=32):
# Verify img_size is a multiple of stride s
new_size = make_divisible(img_size, int(s)) # ceil gs-multiple
if new_size != img_size:
print('WARNING: --img-size %g must be multiple of max stride %g, updating to %g' % (img_size, s, new_size))
return new_size
def check_file(file):
# Search for file if not found
if os.path.isfile(file) or file == '':
return file
else:
files = glob.glob('./**/' + file, recursive=True) # find file
assert len(files), 'File Not Found: %s' % file # assert file was found
assert len(files) == 1, "Multiple files match '%s', specify exact path: %s" % (file, files) # assert unique
return files[0] # return file
def check_dataset(dict):
# Download dataset if not found locally
val, s = dict.get('val'), dict.get('download')
if val and len(val):
val = [Path(x).resolve() for x in (val if isinstance(val, list) else [val])] # val path
if not all(x.exists() for x in val):
print('\nWARNING: Dataset not found, nonexistent paths: %s' % [str(x) for x in val if not x.exists()])
if s and len(s): # download script
print('Downloading %s ...' % s)
if s.startswith('http') and s.endswith('.zip'): # URL
f = Path(s).name # filename
torch.hub.download_url_to_file(s, f)
r = os.system('unzip -q %s -d ../ && rm %s' % (f, f)) # unzip
else: # bash script
r = os.system(s)
print('Dataset autodownload %s\n' % ('success' if r == 0 else 'failure')) # analyze return value
else:
raise Exception('Dataset not found.')
def make_divisible(x, divisor):
# Returns x evenly divisible by divisor
return math.ceil(x / divisor) * divisor
def labels_to_class_weights(labels, nc=80):
# Get class weights (inverse frequency) from training labels
if labels[0] is None: # no labels loaded
return torch.Tensor()
labels = np.concatenate(labels, 0) # labels.shape = (866643, 5) for COCO
classes = labels[:, 0].astype(np.int) # labels = [class xywh]
weights = np.bincount(classes, minlength=nc) # occurrences per class
# Prepend gridpoint count (for uCE training)
# gpi = ((320 / 32 * np.array([1, 2, 4])) ** 2 * 3).sum() # gridpoints per image
# weights = np.hstack([gpi * len(labels) - weights.sum() * 9, weights * 9]) ** 0.5 # prepend gridpoints to start
weights[weights == 0] = 1 # replace empty bins with 1
weights = 1 / weights # number of targets per class
weights /= weights.sum() # normalize
return torch.from_numpy(weights)
def labels_to_image_weights(labels, nc=80, class_weights=np.ones(80)):
# Produces image weights based on class mAPs
n = len(labels)
class_counts = np.array([np.bincount(labels[i][:, 0].astype(np.int), minlength=nc) for i in range(n)])
image_weights = (class_weights.reshape(1, nc) * class_counts).sum(1)
# index = random.choices(range(n), weights=image_weights, k=1) # weight image sample
return image_weights
def coco80_to_coco91_class(): # converts 80-index (val2014) to 91-index (paper)
# https://tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/
# a = np.loadtxt('data/coco.names', dtype='str', delimiter='\n')
# b = np.loadtxt('data/coco_paper.names', dtype='str', delimiter='\n')
# x1 = [list(a[i] == b).index(True) + 1 for i in range(80)] # darknet to coco
# x2 = [list(b[i] == a).index(True) if any(b[i] == a) else None for i in range(91)] # coco to darknet
x = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27, 28, 31, 32, 33, 34,
35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
64, 65, 67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 84, 85, 86, 87, 88, 89, 90]
return x
def xyxy2xywh(x):
# Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] where xy1=top-left, xy2=bottom-right
y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
y[:, 0] = (x[:, 0] + x[:, 2]) / 2 # x center
y[:, 1] = (x[:, 1] + x[:, 3]) / 2 # y center
y[:, 2] = x[:, 2] - x[:, 0] # width
y[:, 3] = x[:, 3] - x[:, 1] # height
return y
def xywh2xyxy(x):
# Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
y[:, 0] = x[:, 0] - x[:, 2] / 2 # top left x
y[:, 1] = x[:, 1] - x[:, 3] / 2 # top left y
y[:, 2] = x[:, 0] + x[:, 2] / 2 # bottom right x
y[:, 3] = x[:, 1] + x[:, 3] / 2 # bottom right y
return y
def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
# Rescale coords (xyxy) from img1_shape to img0_shape
if ratio_pad is None: # calculate from img0_shape
gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1]) # gain = old / new
pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2 # wh padding
else:
gain = ratio_pad[0][0]
pad = ratio_pad[1]
coords[:, [0, 2]] -= pad[0] # x padding
coords[:, [1, 3]] -= pad[1] # y padding
coords[:, :4] /= gain
clip_coords(coords, img0_shape)
return coords
def clip_coords(boxes, img_shape):
# Clip bounding xyxy bounding boxes to image shape (height, width)
boxes[:, 0].clamp_(0, img_shape[1]) # x1
boxes[:, 1].clamp_(0, img_shape[0]) # y1
boxes[:, 2].clamp_(0, img_shape[1]) # x2
boxes[:, 3].clamp_(0, img_shape[0]) # y2
def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-9):
# Returns the IoU of box1 to box2. box1 is 4, box2 is nx4
box2 = box2.T
# Get the coordinates of bounding boxes
if x1y1x2y2: # x1, y1, x2, y2 = box1
b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
else: # transform from xywh to xyxy
b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2
b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2
b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2
b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2
# Intersection area
inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
(torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
# Union Area
w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps
w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps
union = w1 * h1 + w2 * h2 - inter + eps
iou = inter / union
if GIoU or DIoU or CIoU:
cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex (smallest enclosing box) width
ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height
if CIoU or DIoU: # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
c2 = cw ** 2 + ch ** 2 + eps # convex diagonal squared
rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 +
(b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4 # center distance squared
if DIoU:
return iou - rho2 / c2 # DIoU
elif CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
with torch.no_grad():
alpha = v / ((1 + eps) - iou + v)
return iou - (rho2 / c2 + v * alpha) # CIoU
else: # GIoU https://arxiv.org/pdf/1902.09630.pdf
c_area = cw * ch + eps # convex area
return iou - (c_area - union) / c_area # GIoU
else:
return iou # IoU
def box_iou(box1, box2):
# https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
"""
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
box1 (Tensor[N, 4])
box2 (Tensor[M, 4])
Returns:
iou (Tensor[N, M]): the NxM matrix containing the pairwise
IoU values for every element in boxes1 and boxes2
"""
def box_area(box):
# box = 4xn
return (box[2] - box[0]) * (box[3] - box[1])
area1 = box_area(box1.T)
area2 = box_area(box2.T)
# inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
return inter / (area1[:, None] + area2 - inter) # iou = inter / (area1 + area2 - inter)
def wh_iou(wh1, wh2):
# Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
wh1 = wh1[:, None] # [N,1,2]
wh2 = wh2[None] # [1,M,2]
inter = torch.min(wh1, wh2).prod(2) # [N,M]
return inter / (wh1.prod(2) + wh2.prod(2) - inter) # iou = inter / (area1 + area2 - inter)
def non_max_suppression(prediction, conf_thres=0.1, iou_thres=0.6, classes=None, agnostic=False, labels=()):
"""Performs Non-Maximum Suppression (NMS) on inference results
Returns:
detections with shape: nx6 (x1, y1, x2, y2, conf, cls)
"""
nc = prediction[0].shape[1] - 5 # number of classes
xc = prediction[..., 4] > conf_thres # candidates
# Settings
min_wh, max_wh = 2, 4096 # (pixels) minimum and maximum box width and height
max_det = 300 # maximum number of detections per image
time_limit = 10.0 # seconds to quit after
redundant = True # require redundant detections
multi_label = nc > 1 # multiple labels per box (adds 0.5ms/img)
merge = False # use merge-NMS
t = time.time()
output = [torch.zeros(0, 6)] * prediction.shape[0]
for xi, x in enumerate(prediction): # image index, image inference
# Apply constraints
# x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0 # width-height
x = x[xc[xi]] # confidence
# Cat apriori labels if autolabelling
if labels and len(labels[xi]):
l = labels[xi]
v = torch.zeros((len(l), nc + 5), device=x.device)
v[:, :4] = l[:, 1:5] # box
v[:, 4] = 1.0 # conf
v[range(len(l)), l[:, 0].long() + 5] = 1.0 # cls
x = torch.cat((x, v), 0)
# If none remain process next image
if not x.shape[0]:
continue
# Compute conf
x[:, 5:] *= x[:, 4:5] # conf = obj_conf * cls_conf
# Box (center x, center y, width, height) to (x1, y1, x2, y2)
box = xywh2xyxy(x[:, :4])
# Detections matrix nx6 (xyxy, conf, cls)
if multi_label:
i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
else: # best class only
conf, j = x[:, 5:].max(1, keepdim=True)
x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
# Filter by class
if classes:
x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
# Apply finite constraint
# if not torch.isfinite(x).all():
# x = x[torch.isfinite(x).all(1)]
# If none remain process next image
n = x.shape[0] # number of boxes
if not n:
continue
# Sort by confidence
# x = x[x[:, 4].argsort(descending=True)]
# Batched NMS
c = x[:, 5:6] * (0 if agnostic else max_wh) # classes
boxes, scores = x[:, :4] + c, x[:, 4] # boxes (offset by class), scores
i = torchvision.ops.nms(boxes, scores, iou_thres) # NMS
if i.shape[0] > max_det: # limit detections
i = i[:max_det]
if merge and (1 < n < 3E3): # Merge NMS (boxes merged using weighted mean)
# update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
iou = box_iou(boxes[i], boxes) > iou_thres # iou matrix
weights = iou * scores[None] # box weights
x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True) # merged boxes
if redundant:
i = i[iou.sum(1) > 1] # require redundancy
output[xi] = x[i]
if (time.time() - t) > time_limit:
break # time limit exceeded
return output
def strip_optimizer(f='weights/best.pt', s=''): # from utils.general import *; strip_optimizer()
# Strip optimizer from 'f' to finalize training, optionally save as 's'
x = torch.load(f, map_location=torch.device('cpu'))
x['optimizer'] = None
x['training_results'] = None
x['epoch'] = -1
x['model'].half() # to FP16
for p in x['model'].parameters():
p.requires_grad = False
torch.save(x, s or f)
mb = os.path.getsize(s or f) / 1E6 # filesize
print('Optimizer stripped from %s,%s %.1fMB' % (f, (' saved as %s,' % s) if s else '', mb))
def print_mutation(hyp, results, yaml_file='hyp_evolved.yaml', bucket=''):
# Print mutation results to evolve.txt (for use with train.py --evolve)
a = '%10s' * len(hyp) % tuple(hyp.keys()) # hyperparam keys
b = '%10.3g' * len(hyp) % tuple(hyp.values()) # hyperparam values
c = '%10.4g' * len(results) % results # results (P, R, mAP@0.5, mAP@0.5:0.95, val_losses x 3)
print('\n%s\n%s\nEvolved fitness: %s\n' % (a, b, c))
if bucket:
url = 'gs://%s/evolve.txt' % bucket
if gsutil_getsize(url) > (os.path.getsize('evolve.txt') if os.path.exists('evolve.txt') else 0):
os.system('gsutil cp %s .' % url) # download evolve.txt if larger than local
with open('evolve.txt', 'a') as f: # append result
f.write(c + b + '\n')
x = np.unique(np.loadtxt('evolve.txt', ndmin=2), axis=0) # load unique rows
x = x[np.argsort(-fitness(x))] # sort
np.savetxt('evolve.txt', x, '%10.3g') # save sort by fitness
# Save yaml
for i, k in enumerate(hyp.keys()):
hyp[k] = float(x[0, i + 7])
with open(yaml_file, 'w') as f:
results = tuple(x[0, :7])
c = '%10.4g' * len(results) % results # results (P, R, mAP@0.5, mAP@0.5:0.95, val_losses x 3)
f.write('# Hyperparameter Evolution Results\n# Generations: %g\n# Metrics: ' % len(x) + c + '\n\n')
yaml.dump(hyp, f, sort_keys=False)
if bucket:
os.system('gsutil cp evolve.txt %s gs://%s' % (yaml_file, bucket)) # upload
def apply_classifier(x, model, img, im0):
# applies a second stage classifier to yolo outputs
im0 = [im0] if isinstance(im0, np.ndarray) else im0
for i, d in enumerate(x): # per image
if d is not None and len(d):
d = d.clone()
# Reshape and pad cutouts
b = xyxy2xywh(d[:, :4]) # boxes
b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1) # rectangle to square
b[:, 2:] = b[:, 2:] * 1.3 + 30 # pad
d[:, :4] = xywh2xyxy(b).long()
# Rescale boxes from img_size to im0 size
scale_coords(img.shape[2:], d[:, :4], im0[i].shape)
# Classes
pred_cls1 = d[:, 5].long()
ims = []
for j, a in enumerate(d): # per item
cutout = im0[i][int(a[1]):int(a[3]), int(a[0]):int(a[2])]
im = cv2.resize(cutout, (224, 224)) # BGR
# cv2.imwrite('test%i.jpg' % j, cutout)
im = im[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
im = np.ascontiguousarray(im, dtype=np.float32) # uint8 to float32
im /= 255.0 # 0 - 255 to 0.0 - 1.0
ims.append(im)
pred_cls2 = model(torch.Tensor(ims).to(d.device)).argmax(1) # classifier prediction
x[i] = x[i][pred_cls1 == pred_cls2] # retain matching class detections
return x
def increment_path(path, exist_ok=True, sep=''):
# Increment path, i.e. runs/exp --> runs/exp{sep}0, runs/exp{sep}1 etc.
path = Path(path) # os-agnostic
if (path.exists() and exist_ok) or (not path.exists()):
return str(path)
else:
dirs = glob.glob(f"{path}{sep}*") # similar paths
matches = [re.search(rf"%s{sep}(\d+)" % path.stem, d) for d in dirs]
i = [int(m.groups()[0]) for m in matches if m] # indices
n = max(i) + 1 if i else 2 # increment number
return f"{path}{sep}{n}" # update path
utils/google_app_engine/Dockerfile
+25
View File
@@ -0,0 +1,25 @@
FROM gcr.io/google-appengine/python
# Create a virtualenv for dependencies. This isolates these packages from
# system-level packages.
# Use -p python3 or -p python3.7 to select python version. Default is version 2.
RUN virtualenv /env -p python3
# Setting these environment variables are the same as running
# source /env/bin/activate.
ENV VIRTUAL_ENV /env
ENV PATH /env/bin:$PATH
RUN apt-get update && apt-get install -y python-opencv
# Copy the application's requirements.txt and run pip to install all
# dependencies into the virtualenv.
ADD requirements.txt /app/requirements.txt
RUN pip install -r /app/requirements.txt
# Add the application source code.
ADD . /app
# Run a WSGI server to serve the application. gunicorn must be declared as
# a dependency in requirements.txt.
CMD gunicorn -b :$PORT main:app
utils/google_app_engine/additional_requirements.txt
+4
View File
@@ -0,0 +1,4 @@
# add these requirements in your app on top of the existing ones
pip==18.1
Flask==1.0.2
gunicorn==19.9.0
utils/google_app_engine/app.yaml
+14
View File
@@ -0,0 +1,14 @@
runtime: custom
env: flex
service: yolov3app
liveness_check:
initial_delay_sec: 600
manual_scaling:
instances: 1
resources:
cpu: 1
memory_gb: 4
disk_size_gb: 20
utils/google_utils.py
+87 -37
View File
@@ -1,17 +1,60 @@
# This file contains google utils: https://cloud.google.com/storage/docs/reference/libraries
# pip install --upgrade google-cloud-storage
# Google utils: https://cloud.google.com/storage/docs/reference/libraries
import os
import platform
import subprocess
import time
from pathlib import Path
import torch
# from google.cloud import storage
def gsutil_getsize(url=''):
# gs://bucket/file size https://cloud.google.com/storage/docs/gsutil/commands/du
s = subprocess.check_output('gsutil du %s' % url, shell=True).decode('utf-8')
return eval(s.split(' ')[0]) if len(s) else 0 # bytes
def gdrive_download(id='1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO', name='coco.zip'):
# https://gist.github.com/tanaikech/f0f2d122e05bf5f971611258c22c110f
# Downloads a file from Google Drive, accepting presented query
# from utils.google_utils import *; gdrive_download()
def attempt_download(weights):
# Attempt to download pretrained weights if not found locally
weights = weights.strip().replace("'", '')
file = Path(weights).name.lower()
msg = weights + ' missing, try downloading from https://github.com/ultralytics/yolov3/releases/'
models = ['yolov3.pt', 'yolov3-spp.pt', 'yolov3-tiny.pt'] # available models
redundant = False # offer second download option
if file in models and not os.path.isfile(weights):
# Google Drive
# d = {'yolov5s.pt': '1R5T6rIyy3lLwgFXNms8whc-387H0tMQO',
# 'yolov5m.pt': '1vobuEExpWQVpXExsJ2w-Mbf3HJjWkQJr',
# 'yolov5l.pt': '1hrlqD1Wdei7UT4OgT785BEk1JwnSvNEV',
# 'yolov5x.pt': '1mM8aZJlWTxOg7BZJvNUMrTnA2AbeCVzS'}
# r = gdrive_download(id=d[file], name=weights) if file in d else 1
# if r == 0 and os.path.exists(weights) and os.path.getsize(weights) > 1E6: # check
# return
try: # GitHub
url = 'https://github.com/ultralytics/yolov3/releases/download/v1.0/' + file
print('Downloading %s to %s...' % (url, weights))
torch.hub.download_url_to_file(url, weights)
assert os.path.exists(weights) and os.path.getsize(weights) > 1E6 # check
except Exception as e: # GCP
print('Download error: %s' % e)
assert redundant, 'No secondary mirror'
url = 'https://storage.googleapis.com/ultralytics/yolov3/ckpt/' + file
print('Downloading %s to %s...' % (url, weights))
r = os.system('curl -L %s -o %s' % (url, weights)) # torch.hub.download_url_to_file(url, weights)
finally:
if not (os.path.exists(weights) and os.path.getsize(weights) > 1E6): # check
os.remove(weights) if os.path.exists(weights) else None # remove partial downloads
print('ERROR: Download failure: %s' % msg)
print('')
return
def gdrive_download(id='1n_oKgR81BJtqk75b00eAjdv03qVCQn2f', name='coco128.zip'):
# Downloads a file from Google Drive. from utils.google_utils import *; gdrive_download()
t = time.time()
print('Downloading https://drive.google.com/uc?export=download&id=%s as %s... ' % (id, name), end='')
@@ -19,13 +62,13 @@ def gdrive_download(id='1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO', name='coco.zip'):
os.remove('cookie') if os.path.exists('cookie') else None
# Attempt file download
os.system("curl -c ./cookie -s -L \"https://drive.google.com/uc?export=download&id=%s\" > /dev/null" % id)
out = "NUL" if platform.system() == "Windows" else "/dev/null"
os.system('curl -c ./cookie -s -L "drive.google.com/uc?export=download&id=%s" > %s ' % (id, out))
if os.path.exists('cookie'): # large file
s = "curl -Lb ./cookie \"https://drive.google.com/uc?export=download&confirm=`awk '/download/ {print $NF}' ./cookie`&id=%s\" -o %s" % (
id, name)
s = 'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm=%s&id=%s" -o %s' % (get_token(), id, name)
else: # small file
s = "curl -s -L -o %s 'https://drive.google.com/uc?export=download&id=%s'" % (name, id)
r = os.system(s) # execute, capture return values
s = 'curl -s -L -o %s "drive.google.com/uc?export=download&id=%s"' % (name, id)
r = os.system(s) # execute, capture return
os.remove('cookie') if os.path.exists('cookie') else None
# Error check
@@ -44,29 +87,36 @@ def gdrive_download(id='1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO', name='coco.zip'):
return r
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
def get_token(cookie="./cookie"):
with open(cookie) as f:
for line in f:
if "download" in line:
return line.split()[-1]
return ""
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))
# 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))
utils/layers.py
-148
View File
@@ -1,148 +0,0 @@
import torch.nn.functional as F
from utils.utils import *
def make_divisible(v, divisor):
# Function ensures all layers have a channel number that is divisible by 8
# https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
return math.ceil(v / divisor) * divisor
class Flatten(nn.Module):
# Use after nn.AdaptiveAvgPool2d(1) to remove last 2 dimensions
def forward(self, x):
return x.view(x.size(0), -1)
class Concat(nn.Module):
# Concatenate a list of tensors along dimension
def __init__(self, dimension=1):
super(Concat, self).__init__()
self.d = dimension
def forward(self, x):
return torch.cat(x, self.d)
class FeatureConcat(nn.Module):
def __init__(self, layers):
super(FeatureConcat, self).__init__()
self.layers = layers # layer indices
self.multiple = len(layers) > 1 # multiple layers flag
def forward(self, x, outputs):
return torch.cat([outputs[i] for i in self.layers], 1) if self.multiple else outputs[self.layers[0]]
class WeightedFeatureFusion(nn.Module): # weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070
def __init__(self, layers, weight=False):
super(WeightedFeatureFusion, self).__init__()
self.layers = layers # layer indices
self.weight = weight # apply weights boolean
self.n = len(layers) + 1 # number of layers
if weight:
self.w = nn.Parameter(torch.zeros(self.n), requires_grad=True) # layer weights
def forward(self, x, outputs):
# Weights
if self.weight:
w = torch.sigmoid(self.w) * (2 / self.n) # sigmoid weights (0-1)
x = x * w[0]
# Fusion
nx = x.shape[1] # input channels
for i in range(self.n - 1):
a = outputs[self.layers[i]] * w[i + 1] if self.weight else outputs[self.layers[i]] # feature to add
na = a.shape[1] # feature channels
# Adjust channels
if nx == na: # same shape
x = x + a
elif nx > na: # slice input
x[:, :na] = x[:, :na] + a # or a = nn.ZeroPad2d((0, 0, 0, 0, 0, dc))(a); x = x + a
else: # slice feature
x = x + a[:, :nx]
return x
class MixConv2d(nn.Module): # MixConv: Mixed Depthwise Convolutional Kernels https://arxiv.org/abs/1907.09595
def __init__(self, in_ch, out_ch, k=(3, 5, 7), stride=1, dilation=1, bias=True, method='equal_params'):
super(MixConv2d, self).__init__()
groups = len(k)
if method == 'equal_ch': # equal channels per group
i = torch.linspace(0, groups - 1E-6, out_ch).floor() # out_ch indices
ch = [(i == g).sum() for g in range(groups)]
else: # 'equal_params': equal parameter count per group
b = [out_ch] + [0] * groups
a = np.eye(groups + 1, groups, k=-1)
a -= np.roll(a, 1, axis=1)
a *= np.array(k) ** 2
a[0] = 1
ch = np.linalg.lstsq(a, b, rcond=None)[0].round().astype(int) # solve for equal weight indices, ax = b
self.m = nn.ModuleList([nn.Conv2d(in_channels=in_ch,
out_channels=ch[g],
kernel_size=k[g],
stride=stride,
padding=k[g] // 2, # 'same' pad
dilation=dilation,
bias=bias) for g in range(groups)])
def forward(self, x):
return torch.cat([m(x) for m in self.m], 1)
# Activation functions below -------------------------------------------------------------------------------------------
class SwishImplementation(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
return x * torch.sigmoid(x)
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_tensors[0]
sx = torch.sigmoid(x) # sigmoid(ctx)
return grad_output * (sx * (1 + x * (1 - sx)))
class MishImplementation(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x)))
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_tensors[0]
sx = torch.sigmoid(x)
fx = F.softplus(x).tanh()
return grad_output * (fx + x * sx * (1 - fx * fx))
class MemoryEfficientSwish(nn.Module):
def forward(self, x):
return SwishImplementation.apply(x)
class MemoryEfficientMish(nn.Module):
def forward(self, x):
return MishImplementation.apply(x)
class Swish(nn.Module):
def forward(self, x):
return x * torch.sigmoid(x)
class HardSwish(nn.Module): # https://arxiv.org/pdf/1905.02244.pdf
def forward(self, x):
return x * F.hardtanh(x + 3, 0., 6., True) / 6.
class Mish(nn.Module): # https://github.com/digantamisra98/Mish
def forward(self, x):
return x * F.softplus(x).tanh()
utils/loss.py
+179
View File
@@ -0,0 +1,179 @@
# Loss functions
import torch
import torch.nn as nn
from utils.general import bbox_iou
from utils.torch_utils import is_parallel
def smooth_BCE(eps=0.1): # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441
# return positive, negative label smoothing BCE targets
return 1.0 - 0.5 * eps, 0.5 * eps
class BCEBlurWithLogitsLoss(nn.Module):
# BCEwithLogitLoss() with reduced missing label effects.
def __init__(self, alpha=0.05):
super(BCEBlurWithLogitsLoss, self).__init__()
self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none') # must be nn.BCEWithLogitsLoss()
self.alpha = alpha
def forward(self, pred, true):
loss = self.loss_fcn(pred, true)
pred = torch.sigmoid(pred) # prob from logits
dx = pred - true # reduce only missing label effects
# dx = (pred - true).abs() # reduce missing label and false label effects
alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4))
loss *= alpha_factor
return loss.mean()
class FocalLoss(nn.Module):
# Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
super(FocalLoss, self).__init__()
self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
self.gamma = gamma
self.alpha = alpha
self.reduction = loss_fcn.reduction
self.loss_fcn.reduction = 'none' # required to apply FL to each element
def forward(self, pred, true):
loss = self.loss_fcn(pred, true)
# p_t = torch.exp(-loss)
# loss *= self.alpha * (1.000001 - p_t) ** self.gamma # non-zero power for gradient stability
# TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py
pred_prob = torch.sigmoid(pred) # prob from logits
p_t = true * pred_prob + (1 - true) * (1 - pred_prob)
alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
modulating_factor = (1.0 - p_t) ** self.gamma
loss *= alpha_factor * modulating_factor
if self.reduction == 'mean':
return loss.mean()
elif self.reduction == 'sum':
return loss.sum()
else: # 'none'
return loss
def compute_loss(p, targets, model): # predictions, targets, model
device = targets.device
lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
tcls, tbox, indices, anchors = build_targets(p, targets, model) # targets
h = model.hyp # hyperparameters
# Define criteria
BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.Tensor([h['cls_pw']])).to(device)
BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.Tensor([h['obj_pw']])).to(device)
# Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
cp, cn = smooth_BCE(eps=0.0)
# Focal loss
g = h['fl_gamma'] # focal loss gamma
if g > 0:
BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
# Losses
nt = 0 # number of targets
no = len(p) # number of outputs
balance = [4.0, 1.0, 0.4] if no == 3 else [4.0, 1.0, 0.4, 0.1] # P3-5 or P3-6
for i, pi in enumerate(p): # layer index, layer predictions
b, a, gj, gi = indices[i] # image, anchor, gridy, gridx
tobj = torch.zeros_like(pi[..., 0], device=device) # target obj
n = b.shape[0] # number of targets
if n:
nt += n # cumulative targets
ps = pi[b, a, gj, gi] # prediction subset corresponding to targets
# Regression
pxy = ps[:, :2].sigmoid() * 2. - 0.5
pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
pbox = torch.cat((pxy, pwh), 1).to(device) # predicted box
iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target)
lbox += (1.0 - iou).mean() # iou loss
# Objectness
tobj[b, a, gj, gi] = (1.0 - model.gr) + model.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio
# Classification
if model.nc > 1: # cls loss (only if multiple classes)
t = torch.full_like(ps[:, 5:], cn, device=device) # targets
t[range(n), tcls[i]] = cp
lcls += BCEcls(ps[:, 5:], t) # BCE
# Append targets to text file
# with open('targets.txt', 'a') as file:
# [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
lobj += BCEobj(pi[..., 4], tobj) * balance[i] # obj loss
s = 3 / no # output count scaling
lbox *= h['box'] * s
lobj *= h['obj'] * s * (1.4 if no == 4 else 1.)
lcls *= h['cls'] * s
bs = tobj.shape[0] # batch size
loss = lbox + lobj + lcls
return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach()
def build_targets(p, targets, model):
# Build targets for compute_loss(), input targets(image,class,x,y,w,h)
det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module
na, nt = det.na, targets.shape[0] # number of anchors, targets
tcls, tbox, indices, anch = [], [], [], []
gain = torch.ones(7, device=targets.device) # normalized to gridspace gain
ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt)
targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices
g = 0.5 # bias
off = torch.tensor([[0, 0],
# [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
# [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
], device=targets.device).float() * g # offsets
for i in range(det.nl):
anchors = det.anchors[i]
gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain
# Match targets to anchors
t = targets * gain
if nt:
# Matches
r = t[:, :, 4:6] / anchors[:, None] # wh ratio
j = torch.max(r, 1. / r).max(2)[0] < model.hyp['anchor_t'] # compare
# j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
t = t[j] # filter
# Offsets
gxy = t[:, 2:4] # grid xy
gxi = gain[[2, 3]] - gxy # inverse
j, k = ((gxy % 1. < g) & (gxy > 1.)).T
l, m = ((gxi % 1. < g) & (gxi > 1.)).T
j = torch.stack((torch.ones_like(j),))
t = t.repeat((off.shape[0], 1, 1))[j]
offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
else:
t = targets[0]
offsets = 0
# Define
b, c = t[:, :2].long().T # image, class
gxy = t[:, 2:4] # grid xy
gwh = t[:, 4:6] # grid wh
gij = (gxy - offsets).long()
gi, gj = gij.T # grid xy indices
# Append
a = t[:, 6].long() # anchor indices
indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
tbox.append(torch.cat((gxy - gij, gwh), 1)) # box
anch.append(anchors[a]) # anchors
tcls.append(c) # class
return tcls, tbox, indices, anch
utils/metrics.py
+203
View File
@@ -0,0 +1,203 @@
# Model validation metrics
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import torch
from . import general
def fitness(x):
# Model fitness as a weighted combination of metrics
w = [0.0, 0.0, 0.1, 0.9] # weights for [P, R, mAP@0.5, mAP@0.5:0.95]
return (x[:, :4] * w).sum(1)
def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='precision-recall_curve.png', names=[]):
""" Compute the average precision, given the recall and precision curves.
Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
# Arguments
tp: True positives (nparray, nx1 or nx10).
conf: Objectness value from 0-1 (nparray).
pred_cls: Predicted object classes (nparray).
target_cls: True object classes (nparray).
plot: Plot precision-recall curve at mAP@0.5
save_dir: Plot save directory
# Returns
The average precision as computed in py-faster-rcnn.
"""
# Sort by objectness
i = np.argsort(-conf)
tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]
# Find unique classes
unique_classes = np.unique(target_cls)
# Create Precision-Recall curve and compute AP for each class
px, py = np.linspace(0, 1, 1000), [] # for plotting
pr_score = 0.1 # score to evaluate P and R https://github.com/ultralytics/yolov3/issues/898
s = [unique_classes.shape[0], tp.shape[1]] # number class, number iou thresholds (i.e. 10 for mAP0.5...0.95)
ap, p, r = np.zeros(s), np.zeros(s), np.zeros(s)
for ci, c in enumerate(unique_classes):
i = pred_cls == c
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)
# Recall
recall = tpc / (n_l + 1e-16) # recall curve
r[ci] = np.interp(-pr_score, -conf[i], recall[:, 0]) # r at pr_score, negative x, xp because xp decreases
# Precision
precision = tpc / (tpc + fpc) # precision curve
p[ci] = np.interp(-pr_score, -conf[i], precision[:, 0]) # 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
# Compute F1 score (harmonic mean of precision and recall)
f1 = 2 * p * r / (p + r + 1e-16)
if plot:
plot_pr_curve(px, py, ap, save_dir, names)
return p, r, ap, f1, unique_classes.astype('int32')
def compute_ap(recall, precision):
""" Compute the average precision, given the recall and precision curves.
Source: https://github.com/rbgirshick/py-faster-rcnn.
# Arguments
recall: The recall curve (list).
precision: The precision curve (list).
# Returns
The average precision as computed in py-faster-rcnn.
"""
# Append sentinel values to beginning and end
mrec = recall # np.concatenate(([0.], recall, [recall[-1] + 1E-3]))
mpre = precision # np.concatenate(([0.], precision, [0.]))
# Compute the precision envelope
mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))
# Integrate area under curve
method = 'interp' # methods: 'continuous', 'interp'
if method == 'interp':
x = np.linspace(0, 1, 101) # 101-point interp (COCO)
ap = np.trapz(np.interp(x, mrec, mpre), x) # integrate
else: # 'continuous'
i = np.where(mrec[1:] != mrec[:-1])[0] # points where x axis (recall) changes
ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) # area under curve
return ap, mpre, mrec
class ConfusionMatrix:
# Updated version of https://github.com/kaanakan/object_detection_confusion_matrix
def __init__(self, nc, conf=0.25, iou_thres=0.45):
self.matrix = np.zeros((nc + 1, nc + 1))
self.nc = nc # number of classes
self.conf = conf
self.iou_thres = iou_thres
def process_batch(self, detections, labels):
"""
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
detections (Array[N, 6]), x1, y1, x2, y2, conf, class
labels (Array[M, 5]), class, x1, y1, x2, y2
Returns:
None, updates confusion matrix accordingly
"""
detections = detections[detections[:, 4] > self.conf]
gt_classes = labels[:, 0].int()
detection_classes = detections[:, 5].int()
iou = general.box_iou(labels[:, 1:], detections[:, :4])
x = torch.where(iou > self.iou_thres)
if x[0].shape[0]:
matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy()
if x[0].shape[0] > 1:
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
else:
matches = np.zeros((0, 3))
n = matches.shape[0] > 0
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[gc, detection_classes[m1[j]]] += 1 # correct
else:
self.matrix[gc, self.nc] += 1 # background FP
if n:
for i, dc in enumerate(detection_classes):
if not any(m1 == i):
self.matrix[self.nc, dc] += 1 # background FN
def matrix(self):
return self.matrix
def plot(self, save_dir='', names=()):
try:
import seaborn as sn
array = self.matrix / (self.matrix.sum(0).reshape(1, self.nc + 1) + 1E-6) # normalize
array[array < 0.005] = np.nan # don't annotate (would appear as 0.00)
fig = plt.figure(figsize=(12, 9))
sn.set(font_scale=1.0 if self.nc < 50 else 0.8) # for label size
labels = (0 < len(names) < 99) and len(names) == self.nc # apply names to ticklabels
sn.heatmap(array, annot=self.nc < 30, annot_kws={"size": 8}, cmap='Blues', fmt='.2f', square=True,
xticklabels=names + ['background FN'] if labels else "auto",
yticklabels=names + ['background FP'] if labels else "auto").set_facecolor((1, 1, 1))
fig.axes[0].set_xlabel('True')
fig.axes[0].set_ylabel('Predicted')
fig.tight_layout()
fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250)
except Exception as e:
pass
def print(self):
for i in range(self.nc + 1):
print(' '.join(map(str, self.matrix[i])))
# Plots ----------------------------------------------------------------------------------------------------------------
def plot_pr_curve(px, py, ap, save_dir='.', names=()):
fig, ax = plt.subplots(1, 1, figsize=(9, 6))
py = np.stack(py, axis=1)
if 0 < len(names) < 21: # show mAP in legend if < 10 classes
for i, y in enumerate(py.T):
ax.plot(px, y, linewidth=1, label=f'{names[i]} %.3f' % ap[i, 0]) # plot(recall, precision)
else:
ax.plot(px, py, linewidth=1, color='grey') # plot(recall, precision)
ax.plot(px, py.mean(1), linewidth=3, color='blue', label='all classes %.3f mAP@0.5' % ap[:, 0].mean())
ax.set_xlabel('Recall')
ax.set_ylabel('Precision')
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
plt.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
fig.tight_layout()
fig.savefig(Path(save_dir) / 'precision_recall_curve.png', dpi=250)
utils/parse_config.py
-71
View File
@@ -1,71 +0,0 @@
import os
import numpy as np
def parse_model_cfg(path):
# Parse the yolo *.cfg file and return module definitions path may be 'cfg/yolov3.cfg', 'yolov3.cfg', or 'yolov3'
if not path.endswith('.cfg'): # add .cfg suffix if omitted
path += '.cfg'
if not os.path.exists(path) and os.path.exists('cfg' + os.sep + path): # add cfg/ prefix if omitted
path = 'cfg' + os.sep + path
with open(path, 'r') as f:
lines = f.read().split('\n')
lines = [x for x in lines if x and not x.startswith('#')]
lines = [x.rstrip().lstrip() for x in lines] # get rid of fringe whitespaces
mdefs = [] # module definitions
for line in lines:
if line.startswith('['): # This marks the start of a new block
mdefs.append({})
mdefs[-1]['type'] = line[1:-1].rstrip()
if mdefs[-1]['type'] == 'convolutional':
mdefs[-1]['batch_normalize'] = 0 # pre-populate with zeros (may be overwritten later)
else:
key, val = line.split("=")
key = key.rstrip()
if key == 'anchors': # return nparray
mdefs[-1][key] = np.array([float(x) for x in val.split(',')]).reshape((-1, 2)) # np anchors
elif (key in ['from', 'layers', 'mask']) or (key == 'size' and ',' in val): # return array
mdefs[-1][key] = [int(x) for x in val.split(',')]
else:
val = val.strip()
# TODO: .isnumeric() actually fails to get the float case
if val.isnumeric(): # return int or float
mdefs[-1][key] = int(val) if (int(val) - float(val)) == 0 else float(val)
else:
mdefs[-1][key] = val # return string
# Check all fields are supported
supported = ['type', 'batch_normalize', 'filters', 'size', 'stride', 'pad', 'activation', 'layers', 'groups',
'from', 'mask', 'anchors', 'classes', 'num', 'jitter', 'ignore_thresh', 'truth_thresh', 'random',
'stride_x', 'stride_y', 'weights_type', 'weights_normalization', 'scale_x_y', 'beta_nms', 'nms_kind',
'iou_loss', 'iou_normalizer', 'cls_normalizer', 'iou_thresh', 'probability']
f = [] # fields
for x in mdefs[1:]:
[f.append(k) for k in x if k not in f]
u = [x for x in f if x not in supported] # unsupported fields
assert not any(u), "Unsupported fields %s in %s. See https://github.com/ultralytics/yolov3/issues/631" % (u, path)
return mdefs
def parse_data_cfg(path):
# Parses the data configuration file
if not os.path.exists(path) and os.path.exists('data' + os.sep + path): # add data/ prefix if omitted
path = 'data' + os.sep + path
with open(path, 'r') as f:
lines = f.readlines()
options = dict()
for line in lines:
line = line.strip()
if line == '' or line.startswith('#'):
continue
key, val = line.split('=')
options[key.strip()] = val.strip()
return options
utils/plots.py
+379
View File
@@ -0,0 +1,379 @@
# Plotting utils
import glob
import os
import random
from copy import copy
from pathlib import Path
import cv2
import math
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import torch
import yaml
from PIL import Image, ImageDraw
from scipy.signal import butter, filtfilt
from utils.general import xywh2xyxy, xyxy2xywh
from utils.metrics import fitness
# Settings
matplotlib.use('Agg') # for writing to files only
def color_list():
# Return first 10 plt colors as (r,g,b) https://stackoverflow.com/questions/51350872/python-from-color-name-to-rgb
def hex2rgb(h):
return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))
return [hex2rgb(h) for h in plt.rcParams['axes.prop_cycle'].by_key()['color']]
def hist2d(x, y, n=100):
# 2d histogram used in labels.png and evolve.png
xedges, yedges = np.linspace(x.min(), x.max(), n), np.linspace(y.min(), y.max(), n)
hist, xedges, yedges = np.histogram2d(x, y, (xedges, yedges))
xidx = np.clip(np.digitize(x, xedges) - 1, 0, hist.shape[0] - 1)
yidx = np.clip(np.digitize(y, yedges) - 1, 0, hist.shape[1] - 1)
return np.log(hist[xidx, yidx])
def butter_lowpass_filtfilt(data, cutoff=1500, fs=50000, order=5):
# https://stackoverflow.com/questions/28536191/how-to-filter-smooth-with-scipy-numpy
def butter_lowpass(cutoff, fs, order):
nyq = 0.5 * fs
normal_cutoff = cutoff / nyq
return butter(order, normal_cutoff, btype='low', analog=False)
b, a = butter_lowpass(cutoff, fs, order=order)
return filtfilt(b, a, data) # forward-backward filter
def plot_one_box(x, img, color=None, label=None, line_thickness=None):
# Plots one bounding box on image img
tl = line_thickness or round(0.002 * (img.shape[0] + img.shape[1]) / 2) + 1 # line/font thickness
color = color or [random.randint(0, 255) for _ in range(3)]
c1, c2 = (int(x[0]), int(x[1])), (int(x[2]), int(x[3]))
cv2.rectangle(img, c1, c2, color, thickness=tl, lineType=cv2.LINE_AA)
if label:
tf = max(tl - 1, 1) # font thickness
t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3
cv2.rectangle(img, c1, c2, color, -1, cv2.LINE_AA) # filled
cv2.putText(img, label, (c1[0], c1[1] - 2), 0, tl / 3, [225, 255, 255], thickness=tf, lineType=cv2.LINE_AA)
def plot_wh_methods(): # from utils.plots import *; plot_wh_methods()
# Compares the two methods for width-height anchor multiplication
# https://github.com/ultralytics/yolov3/issues/168
x = np.arange(-4.0, 4.0, .1)
ya = np.exp(x)
yb = torch.sigmoid(torch.from_numpy(x)).numpy() * 2
fig = plt.figure(figsize=(6, 3), dpi=150)
plt.plot(x, ya, '.-', label='YOLOv3')
plt.plot(x, yb ** 2, '.-', label='YOLOv5 ^2')
plt.plot(x, yb ** 1.6, '.-', label='YOLOv5 ^1.6')
plt.xlim(left=-4, right=4)
plt.ylim(bottom=0, top=6)
plt.xlabel('input')
plt.ylabel('output')
plt.grid()
plt.legend()
fig.tight_layout()
fig.savefig('comparison.png', dpi=200)
def output_to_target(output):
# Convert model output to target format [batch_id, class_id, x, y, w, h, conf]
targets = []
for i, o in enumerate(output):
for *box, conf, cls in o.cpu().numpy():
targets.append([i, cls, *list(*xyxy2xywh(np.array(box)[None])), conf])
return np.array(targets)
def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=640, max_subplots=16):
# Plot image grid with labels
if isinstance(images, torch.Tensor):
images = images.cpu().float().numpy()
if isinstance(targets, torch.Tensor):
targets = targets.cpu().numpy()
# un-normalise
if np.max(images[0]) <= 1:
images *= 255
tl = 3 # line thickness
tf = max(tl - 1, 1) # font thickness
bs, _, h, w = images.shape # batch size, _, height, width
bs = min(bs, max_subplots) # limit plot images
ns = np.ceil(bs ** 0.5) # number of subplots (square)
# Check if we should resize
scale_factor = max_size / max(h, w)
if scale_factor < 1:
h = math.ceil(scale_factor * h)
w = math.ceil(scale_factor * w)
colors = color_list() # list of colors
mosaic = np.full((int(ns * h), int(ns * w), 3), 255, dtype=np.uint8) # init
for i, img in enumerate(images):
if i == max_subplots: # if last batch has fewer images than we expect
break
block_x = int(w * (i // ns))
block_y = int(h * (i % ns))
img = img.transpose(1, 2, 0)
if scale_factor < 1:
img = cv2.resize(img, (w, h))
mosaic[block_y:block_y + h, block_x:block_x + w, :] = img
if len(targets) > 0:
image_targets = targets[targets[:, 0] == i]
boxes = xywh2xyxy(image_targets[:, 2:6]).T
classes = image_targets[:, 1].astype('int')
labels = image_targets.shape[1] == 6 # labels if no conf column
conf = None if labels else image_targets[:, 6] # check for confidence presence (label vs pred)
if boxes.shape[1] and boxes.max() <= 1: # if normalized
boxes[[0, 2]] *= w # scale to pixels
boxes[[1, 3]] *= h
boxes[[0, 2]] += block_x
boxes[[1, 3]] += block_y
for j, box in enumerate(boxes.T):
cls = int(classes[j])
color = colors[cls % len(colors)]
cls = names[cls] if names else cls
if labels or conf[j] > 0.25: # 0.25 conf thresh
label = '%s' % cls if labels else '%s %.1f' % (cls, conf[j])
plot_one_box(box, mosaic, label=label, color=color, line_thickness=tl)
# Draw image filename labels
if paths:
label = Path(paths[i]).name[:40] # trim to 40 char
t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
cv2.putText(mosaic, label, (block_x + 5, block_y + t_size[1] + 5), 0, tl / 3, [220, 220, 220], thickness=tf,
lineType=cv2.LINE_AA)
# Image border
cv2.rectangle(mosaic, (block_x, block_y), (block_x + w, block_y + h), (255, 255, 255), thickness=3)
if fname:
r = min(1280. / max(h, w) / ns, 1.0) # ratio to limit image size
mosaic = cv2.resize(mosaic, (int(ns * w * r), int(ns * h * r)), interpolation=cv2.INTER_AREA)
# cv2.imwrite(fname, cv2.cvtColor(mosaic, cv2.COLOR_BGR2RGB)) # cv2 save
Image.fromarray(mosaic).save(fname) # PIL save
return mosaic
def plot_lr_scheduler(optimizer, scheduler, epochs=300, save_dir=''):
# Plot LR simulating training for full epochs
optimizer, scheduler = copy(optimizer), copy(scheduler) # do not modify originals
y = []
for _ in range(epochs):
scheduler.step()
y.append(optimizer.param_groups[0]['lr'])
plt.plot(y, '.-', label='LR')
plt.xlabel('epoch')
plt.ylabel('LR')
plt.grid()
plt.xlim(0, epochs)
plt.ylim(0)
plt.tight_layout()
plt.savefig(Path(save_dir) / 'LR.png', dpi=200)
def plot_test_txt(): # from utils.plots import *; plot_test()
# Plot test.txt histograms
x = np.loadtxt('test.txt', dtype=np.float32)
box = xyxy2xywh(x[:, :4])
cx, cy = box[:, 0], box[:, 1]
fig, ax = plt.subplots(1, 1, figsize=(6, 6), tight_layout=True)
ax.hist2d(cx, cy, bins=600, cmax=10, cmin=0)
ax.set_aspect('equal')
plt.savefig('hist2d.png', dpi=300)
fig, ax = plt.subplots(1, 2, figsize=(12, 6), tight_layout=True)
ax[0].hist(cx, bins=600)
ax[1].hist(cy, bins=600)
plt.savefig('hist1d.png', dpi=200)
def plot_targets_txt(): # from utils.plots import *; plot_targets_txt()
# Plot targets.txt histograms
x = np.loadtxt('targets.txt', dtype=np.float32).T
s = ['x targets', 'y targets', 'width targets', 'height targets']
fig, ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)
ax = ax.ravel()
for i in range(4):
ax[i].hist(x[i], bins=100, label='%.3g +/- %.3g' % (x[i].mean(), x[i].std()))
ax[i].legend()
ax[i].set_title(s[i])
plt.savefig('targets.jpg', dpi=200)
def plot_study_txt(path='', x=None): # from utils.plots import *; plot_study_txt()
# Plot study.txt generated by test.py
fig, ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)
ax = ax.ravel()
fig2, ax2 = plt.subplots(1, 1, figsize=(8, 4), tight_layout=True)
for f in [Path(path) / f'study_coco_{x}.txt' for x in ['yolov3', 'yolov3-spp', 'yolov3-tiny']]:
y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
x = np.arange(y.shape[1]) if x is None else np.array(x)
s = ['P', 'R', 'mAP@.5', 'mAP@.5:.95', 't_inference (ms/img)', 't_NMS (ms/img)', 't_total (ms/img)']
for i in range(7):
ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
ax[i].set_title(s[i])
j = y[3].argmax() + 1
ax2.plot(y[6, :j], y[3, :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')
ax2.grid()
ax2.set_xlim(0, 30)
ax2.set_ylim(15, 50)
ax2.set_yticks(np.arange(15, 55, 5))
ax2.set_xlabel('GPU Speed (ms/img)')
ax2.set_ylabel('COCO AP val')
ax2.legend(loc='lower right')
plt.savefig('test_study.png', dpi=300)
def plot_labels(labels, save_dir=''):
# plot dataset labels
c, b = labels[:, 0], labels[:, 1:].transpose() # classes, boxes
nc = int(c.max() + 1) # number of classes
colors = color_list()
# seaborn correlogram
try:
import seaborn as sns
import pandas as pd
x = pd.DataFrame(b.transpose(), columns=['x', 'y', 'width', 'height'])
sns.pairplot(x, corner=True, diag_kind='hist', kind='scatter', markers='o',
plot_kws=dict(s=3, edgecolor=None, linewidth=1, alpha=0.02),
diag_kws=dict(bins=50))
plt.savefig(Path(save_dir) / 'labels_correlogram.png', dpi=200)
plt.close()
except Exception as e:
pass
# matplotlib labels
ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)[1].ravel()
ax[0].hist(c, bins=np.linspace(0, nc, nc + 1) - 0.5, rwidth=0.8)
ax[0].set_xlabel('classes')
ax[2].scatter(b[0], b[1], c=hist2d(b[0], b[1], 90), cmap='jet')
ax[2].set_xlabel('x')
ax[2].set_ylabel('y')
ax[3].scatter(b[2], b[3], c=hist2d(b[2], b[3], 90), cmap='jet')
ax[3].set_xlabel('width')
ax[3].set_ylabel('height')
# rectangles
labels[:, 1:3] = 0.5 # center
labels[:, 1:] = xywh2xyxy(labels[:, 1:]) * 2000
img = Image.fromarray(np.ones((2000, 2000, 3), dtype=np.uint8) * 255)
for cls, *box in labels[:1000]:
ImageDraw.Draw(img).rectangle(box, width=1, outline=colors[int(cls) % 10]) # plot
ax[1].imshow(img)
ax[1].axis('off')
for a in [0, 1, 2, 3]:
for s in ['top', 'right', 'left', 'bottom']:
ax[a].spines[s].set_visible(False)
plt.savefig(Path(save_dir) / 'labels.png', dpi=200)
plt.close()
def plot_evolution(yaml_file='data/hyp.finetune.yaml'): # from utils.plots import *; plot_evolution()
# Plot hyperparameter evolution results in evolve.txt
with open(yaml_file) as f:
hyp = yaml.load(f, Loader=yaml.FullLoader)
x = np.loadtxt('evolve.txt', ndmin=2)
f = fitness(x)
# weights = (f - f.min()) ** 2 # for weighted results
plt.figure(figsize=(10, 12), tight_layout=True)
matplotlib.rc('font', **{'size': 8})
for i, (k, v) in enumerate(hyp.items()):
y = x[:, i + 7]
# mu = (y * weights).sum() / weights.sum() # best weighted result
mu = y[f.argmax()] # best single result
plt.subplot(6, 5, i + 1)
plt.scatter(y, f, c=hist2d(y, f, 20), cmap='viridis', alpha=.8, edgecolors='none')
plt.plot(mu, f.max(), 'k+', markersize=15)
plt.title('%s = %.3g' % (k, mu), fontdict={'size': 9}) # limit to 40 characters
if i % 5 != 0:
plt.yticks([])
print('%15s: %.3g' % (k, mu))
plt.savefig('evolve.png', dpi=200)
print('\nPlot saved as evolve.png')
def plot_results_overlay(start=0, stop=0): # from utils.plots import *; plot_results_overlay()
# Plot training 'results*.txt', overlaying train and val losses
s = ['train', 'train', 'train', 'Precision', 'mAP@0.5', 'val', 'val', 'val', 'Recall', 'mAP@0.5:0.95'] # legends
t = ['Box', 'Objectness', 'Classification', 'P-R', 'mAP-F1'] # titles
for f in sorted(glob.glob('results*.txt') + glob.glob('../../Downloads/results*.txt')):
results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
n = results.shape[1] # number of rows
x = range(start, min(stop, n) if stop else n)
fig, ax = plt.subplots(1, 5, figsize=(14, 3.5), tight_layout=True)
ax = ax.ravel()
for i in range(5):
for j in [i, i + 5]:
y = results[j, x]
ax[i].plot(x, y, marker='.', label=s[j])
# y_smooth = butter_lowpass_filtfilt(y)
# ax[i].plot(x, np.gradient(y_smooth), marker='.', label=s[j])
ax[i].set_title(t[i])
ax[i].legend()
ax[i].set_ylabel(f) if i == 0 else None # add filename
fig.savefig(f.replace('.txt', '.png'), dpi=200)
def plot_results(start=0, stop=0, bucket='', id=(), labels=(), save_dir=''):
# Plot training 'results*.txt'. from utils.plots import *; plot_results(save_dir='runs/train/exp')
fig, ax = plt.subplots(2, 5, figsize=(12, 6))
ax = ax.ravel()
s = ['Box', 'Objectness', 'Classification', 'Precision', 'Recall',
'val Box', 'val Objectness', 'val Classification', 'mAP@0.5', 'mAP@0.5:0.95']
if bucket:
# files = ['https://storage.googleapis.com/%s/results%g.txt' % (bucket, x) for x in id]
files = ['results%g.txt' % x for x in id]
c = ('gsutil cp ' + '%s ' * len(files) + '.') % tuple('gs://%s/results%g.txt' % (bucket, x) for x in id)
os.system(c)
else:
files = list(Path(save_dir).glob('results*.txt'))
assert len(files), 'No results.txt files found in %s, nothing to plot.' % os.path.abspath(save_dir)
for fi, f in enumerate(files):
try:
results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
n = results.shape[1] # number of rows
x = range(start, min(stop, n) if stop else n)
for i in range(10):
y = results[i, x]
if i in [0, 1, 2, 5, 6, 7]:
y[y == 0] = np.nan # don't show zero loss values
# y /= y[0] # normalize
label = labels[fi] if len(labels) else f.stem
ax[i].plot(x, y, marker='.', label=label, linewidth=2, markersize=8)
ax[i].set_title(s[i])
# if i in [5, 6, 7]: # share train and val loss y axes
# ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])
except Exception as e:
print('Warning: Plotting error for %s; %s' % (f, e))
fig.tight_layout()
ax[1].legend()
fig.savefig(Path(save_dir) / 'results.png', dpi=200)
utils/torch_utils.py
+133 -85
View File
@@ -1,24 +1,45 @@
# PyTorch utils
import logging
import math
import os
import time
from contextlib import contextmanager
from copy import deepcopy
import torch
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torch.nn.functional as F
import torchvision
logger = logging.getLogger(__name__)
def init_seeds(seed=0):
@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.
"""
if local_rank not in [-1, 0]:
torch.distributed.barrier()
yield
if local_rank == 0:
torch.distributed.barrier()
def init_torch_seeds(seed=0):
# Speed-reproducibility tradeoff https://pytorch.org/docs/stable/notes/randomness.html
torch.manual_seed(seed)
# Reduce randomness (may be slower on Tesla GPUs) # https://pytorch.org/docs/stable/notes/randomness.html
if seed == 0:
if seed == 0: # slower, more reproducible
cudnn.deterministic = True
cudnn.benchmark = False
else: # faster, less reproducible
cudnn.deterministic = False
cudnn.benchmark = True
def select_device(device='', apex=False, batch_size=None):
def select_device(device='', batch_size=None):
# device = 'cpu' or '0' or '0,1,2,3'
cpu_request = device.lower() == 'cpu'
if device and not cpu_request: # if device requested other than 'cpu'
@@ -32,16 +53,15 @@ def select_device(device='', apex=False, batch_size=None):
if ng > 1 and batch_size: # check that batch_size is compatible with device_count
assert batch_size % ng == 0, 'batch-size %g not multiple of GPU count %g' % (batch_size, ng)
x = [torch.cuda.get_device_properties(i) for i in range(ng)]
s = 'Using CUDA ' + ('Apex ' if apex else '') # apex for mixed precision https://github.com/NVIDIA/apex
s = f'Using torch {torch.__version__} '
for i in range(0, ng):
if i == 1:
s = ' ' * len(s)
print("%sdevice%g _CudaDeviceProperties(name='%s', total_memory=%dMB)" %
(s, i, x[i].name, x[i].total_memory / c))
logger.info("%sCUDA:%g (%s, %dMB)" % (s, i, x[i].name, x[i].total_memory / c))
else:
print('Using CPU')
logger.info(f'Using torch {torch.__version__} CPU')
print('') # skip a line
logger.info('') # skip a line
return torch.device('cuda:0' if cuda else 'cpu')
@@ -50,52 +70,77 @@ def time_synchronized():
return time.time()
def is_parallel(model):
return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel)
def intersect_dicts(da, db, exclude=()):
# Dictionary intersection of matching keys and shapes, omitting 'exclude' keys, using da values
return {k: v for k, v in da.items() if k in db and not any(x in k for x in exclude) and v.shape == db[k].shape}
def initialize_weights(model):
for m in model.modules():
t = type(m)
if t is nn.Conv2d:
pass # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif t is nn.BatchNorm2d:
m.eps = 1e-4
m.eps = 1e-3
m.momentum = 0.03
elif t in [nn.LeakyReLU, nn.ReLU, nn.ReLU6]:
elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6]:
m.inplace = True
def find_modules(model, mclass=nn.Conv2d):
# finds layer indices matching module class 'mclass'
# Finds layer indices matching module class 'mclass'
return [i for i, m in enumerate(model.module_list) if isinstance(m, mclass)]
def sparsity(model):
# Return global model sparsity
a, b = 0., 0.
for p in model.parameters():
a += p.numel()
b += (p == 0).sum()
return b / a
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
print(' %.3g global sparsity' % sparsity(model))
def fuse_conv_and_bn(conv, bn):
# https://tehnokv.com/posts/fusing-batchnorm-and-conv/
with torch.no_grad():
# init
fusedconv = torch.nn.Conv2d(conv.in_channels,
conv.out_channels,
kernel_size=conv.kernel_size,
stride=conv.stride,
padding=conv.padding,
bias=True)
# 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,
groups=conv.groups,
bias=True).requires_grad_(False).to(conv.weight.device)
# 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.size()))
# 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.size()))
# prepare spatial bias
if conv.bias is not None:
b_conv = conv.bias
else:
b_conv = torch.zeros(conv.weight.size(0))
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)
# 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)
return fusedconv
return fusedconv
def model_info(model, verbose=False):
# Plots a line-by-line description of a PyTorch model
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
if verbose:
@@ -107,40 +152,57 @@ def model_info(model, verbose=False):
try: # FLOPS
from thop import profile
macs, _ = profile(model, inputs=(torch.zeros(1, 3, 480, 640),), verbose=False)
fs = ', %.1f GFLOPS' % (macs / 1E9 * 2)
except:
stride = int(model.stride.max())
img = torch.zeros((1, 3, stride, stride), device=next(model.parameters()).device) # input
flops = profile(deepcopy(model), inputs=(img,), verbose=False)[0] / 1E9 * 2 # stride FLOPS
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 FLOPS
except (ImportError, Exception):
fs = ''
print('Model Summary: %g layers, %g parameters, %g gradients%s' % (len(list(model.parameters())), n_p, n_g, fs))
logger.info(f"Model Summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}")
def load_classifier(name='resnet101', n=2):
# Loads a pretrained model reshaped to n-class output
import pretrainedmodels # https://github.com/Cadene/pretrained-models.pytorch#torchvision
model = pretrainedmodels.__dict__[name](num_classes=1000, pretrained='imagenet')
model = torchvision.models.__dict__[name](pretrained=True)
# Display model properties
for x in ['model.input_size', 'model.input_space', 'model.input_range', 'model.mean', 'model.std']:
print(x + ' =', eval(x))
# ResNet model properties
# input_size = [3, 224, 224]
# input_space = 'RGB'
# input_range = [0, 1]
# mean = [0.485, 0.456, 0.406]
# std = [0.229, 0.224, 0.225]
# Reshape output to n classes
filters = model.last_linear.weight.shape[1]
model.last_linear.bias = torch.nn.Parameter(torch.zeros(n))
model.last_linear.weight = torch.nn.Parameter(torch.zeros(n, filters))
model.last_linear.out_features = n
filters = model.fc.weight.shape[1]
model.fc.bias = nn.Parameter(torch.zeros(n), requires_grad=True)
model.fc.weight = nn.Parameter(torch.zeros(n, filters), requires_grad=True)
model.fc.out_features = n
return model
def scale_img(img, ratio=1.0, same_shape=True): # img(16,3,256,416), r=ratio
def scale_img(img, ratio=1.0, same_shape=False): # img(16,3,256,416), r=ratio
# scales img(bs,3,y,x) by ratio
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
gs = 64 # (pixels) grid size
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
if ratio == 1.0:
return img
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
gs = 32 # (pixels) grid size
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=()):
# Copy attributes from b to a, options to only include [...] and to exclude [...]
for k, v in b.__dict__.items():
if (len(include) and k not in include) or k.startswith('_') or k in exclude:
continue
else:
setattr(a, k, v)
class ModelEMA:
@@ -149,46 +211,32 @@ class ModelEMA:
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.
E.g. Google's hyper-params for training MNASNet, MobileNet-V3, EfficientNet, etc that use
RMSprop with a short 2.4-3 epoch decay period and slow LR decay rate of .96-.99 requires EMA
smoothing of weights to match results. Pay attention to the decay constant you are using
relative to your update count per epoch.
To keep EMA from using GPU resources, set device='cpu'. This will save a bit of memory but
disable validation of the EMA weights. Validation will have to be done manually in a separate
process, or after the training stops converging.
This class is sensitive where it is initialized in the sequence of model init,
GPU assignment and distributed training wrappers.
I've tested with the sequence in my own train.py for torch.DataParallel, apex.DDP, and single-GPU.
"""
def __init__(self, model, decay=0.9999, device=''):
# make a copy of the model for accumulating moving average of weights
self.ema = deepcopy(model)
self.ema.eval()
self.updates = 0 # number of EMA updates
def __init__(self, model, decay=0.9999, updates=0):
# Create 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 / 2000)) # decay exponential ramp (to help early epochs)
self.device = device # perform ema on different device from model if set
if device:
self.ema.to(device=device)
for p in self.ema.parameters():
p.requires_grad_(False)
def update(self, model):
self.updates += 1
d = self.decay(self.updates)
# Update EMA parameters
with torch.no_grad():
if type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel):
msd, esd = model.module.state_dict(), self.ema.module.state_dict()
else:
msd, esd = model.state_dict(), self.ema.state_dict()
self.updates += 1
d = self.decay(self.updates)
for k, v in esd.items():
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):
# Assign attributes (which may change during training)
for k in model.__dict__.keys():
if not k.startswith('_'):
setattr(self.ema, k, getattr(model, k))
def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
# Update EMA attributes
copy_attr(self.ema, model, include, exclude)
utils/utils.py
-1080
View File
File diff suppressed because it is too large Load Diff