|
import argparse |
|
|
|
import torch.distributed as dist |
|
import torch.nn.functional as F |
|
import torch.optim as optim |
|
import torch.optim.lr_scheduler as lr_scheduler |
|
import yaml |
|
from torch.utils.tensorboard import SummaryWriter |
|
|
|
import test |
|
from models.yolo import Model |
|
from utils.datasets import * |
|
from utils.utils import * |
|
|
|
mixed_precision = True |
|
try: |
|
from apex import amp |
|
except: |
|
print('Apex recommended for faster mixed precision training: https://github.com/NVIDIA/apex') |
|
mixed_precision = False |
|
|
|
wdir = 'weights' + os.sep |
|
os.makedirs(wdir, exist_ok=True) |
|
last = wdir + 'last.pt' |
|
best = wdir + 'best.pt' |
|
results_file = 'results.txt' |
|
|
|
|
|
hyp = {'lr0': 0.01, |
|
'momentum': 0.937, |
|
'weight_decay': 5e-4, |
|
'giou': 0.05, |
|
'cls': 0.58, |
|
'cls_pw': 1.0, |
|
'obj': 1.0, |
|
'obj_pw': 1.0, |
|
'iou_t': 0.20, |
|
'anchor_t': 4.0, |
|
'fl_gamma': 0.0, |
|
'hsv_h': 0.014, |
|
'hsv_s': 0.68, |
|
'hsv_v': 0.36, |
|
'degrees': 0.0, |
|
'translate': 0.0, |
|
'scale': 0.5, |
|
'shear': 0.0} |
|
print(hyp) |
|
|
|
|
|
f = glob.glob('hyp*.txt') |
|
if f: |
|
print('Using %s' % f[0]) |
|
for k, v in zip(hyp.keys(), np.loadtxt(f[0])): |
|
hyp[k] = v |
|
|
|
|
|
if hyp['fl_gamma']: |
|
print('Using FocalLoss(gamma=%g)' % hyp['fl_gamma']) |
|
|
|
|
|
def train(hyp): |
|
epochs = opt.epochs |
|
batch_size = opt.batch_size |
|
weights = opt.weights |
|
|
|
|
|
init_seeds(1) |
|
with open(opt.data) as f: |
|
data_dict = yaml.load(f, Loader=yaml.FullLoader) |
|
train_path = data_dict['train'] |
|
test_path = data_dict['val'] |
|
nc = 1 if opt.single_cls else int(data_dict['nc']) |
|
|
|
|
|
for f in glob.glob('*_batch*.jpg') + glob.glob(results_file): |
|
os.remove(f) |
|
|
|
|
|
model = Model(opt.cfg).to(device) |
|
assert model.md['nc'] == nc, '%s nc=%g classes but %s nc=%g classes' % (opt.data, nc, opt.cfg, model.md['nc']) |
|
|
|
|
|
gs = int(max(model.stride)) |
|
imgsz, imgsz_test = [check_img_size(x, gs) for x in opt.img_size] |
|
|
|
|
|
nbs = 64 |
|
accumulate = max(round(nbs / batch_size), 1) |
|
hyp['weight_decay'] *= batch_size * accumulate / nbs |
|
pg0, pg1, pg2 = [], [], [] |
|
for k, v in model.named_parameters(): |
|
if v.requires_grad: |
|
if '.bias' in k: |
|
pg2.append(v) |
|
elif '.weight' in k and '.bn' not in k: |
|
pg1.append(v) |
|
else: |
|
pg0.append(v) |
|
|
|
optimizer = optim.Adam(pg0, lr=hyp['lr0']) if opt.adam else \ |
|
optim.SGD(pg0, lr=hyp['lr0'], momentum=hyp['momentum'], nesterov=True) |
|
optimizer.add_param_group({'params': pg1, 'weight_decay': hyp['weight_decay']}) |
|
optimizer.add_param_group({'params': pg2}) |
|
print('Optimizer groups: %g .bias, %g conv.weight, %g other' % (len(pg2), len(pg1), len(pg0))) |
|
del pg0, pg1, pg2 |
|
|
|
|
|
google_utils.attempt_download(weights) |
|
start_epoch, best_fitness = 0, 0.0 |
|
if weights.endswith('.pt'): |
|
ckpt = torch.load(weights, map_location=device) |
|
|
|
|
|
try: |
|
ckpt['model'] = \ |
|
{k: v for k, v in ckpt['model'].state_dict().items() if model.state_dict()[k].numel() == v.numel()} |
|
model.load_state_dict(ckpt['model'], strict=False) |
|
except KeyError as e: |
|
s = "%s is not compatible with %s. Specify --weights '' or specify a --cfg compatible with %s." \ |
|
% (opt.weights, opt.cfg, opt.weights) |
|
raise KeyError(s) from e |
|
|
|
|
|
if ckpt['optimizer'] is not None: |
|
optimizer.load_state_dict(ckpt['optimizer']) |
|
best_fitness = ckpt['best_fitness'] |
|
|
|
|
|
if ckpt.get('training_results') is not None: |
|
with open(results_file, 'w') as file: |
|
file.write(ckpt['training_results']) |
|
|
|
start_epoch = ckpt['epoch'] + 1 |
|
del ckpt |
|
|
|
|
|
if mixed_precision: |
|
model, optimizer = amp.initialize(model, optimizer, opt_level='O1', verbosity=0) |
|
|
|
|
|
lf = lambda x: (((1 + math.cos(x * math.pi / epochs)) / 2) ** 1.0) * 0.9 + 0.1 |
|
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf) |
|
scheduler.last_epoch = start_epoch - 1 |
|
|
|
|
|
|
|
|
|
if device.type != 'cpu' and torch.cuda.device_count() > 1 and torch.distributed.is_available(): |
|
dist.init_process_group(backend='nccl', |
|
init_method='tcp://127.0.0.1:9999', |
|
world_size=1, |
|
rank=0) |
|
model = torch.nn.parallel.DistributedDataParallel(model) |
|
|
|
|
|
dataset = LoadImagesAndLabels(train_path, imgsz, batch_size, |
|
augment=True, |
|
hyp=hyp, |
|
rect=opt.rect, |
|
cache_images=opt.cache_images, |
|
single_cls=opt.single_cls) |
|
mlc = np.concatenate(dataset.labels, 0)[:, 0].max() |
|
assert mlc < nc, 'Label class %g exceeds nc=%g in %s. Correct your labels or your model.' % (mlc, nc, opt.cfg) |
|
|
|
|
|
batch_size = min(batch_size, len(dataset)) |
|
nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8]) |
|
dataloader = torch.utils.data.DataLoader(dataset, |
|
batch_size=batch_size, |
|
num_workers=nw, |
|
shuffle=not opt.rect, |
|
pin_memory=True, |
|
collate_fn=dataset.collate_fn) |
|
|
|
|
|
testloader = torch.utils.data.DataLoader(LoadImagesAndLabels(test_path, imgsz_test, batch_size, |
|
hyp=hyp, |
|
rect=True, |
|
cache_images=opt.cache_images, |
|
single_cls=opt.single_cls), |
|
batch_size=batch_size, |
|
num_workers=nw, |
|
pin_memory=True, |
|
collate_fn=dataset.collate_fn) |
|
|
|
|
|
hyp['cls'] *= nc / 80. |
|
model.nc = nc |
|
model.hyp = hyp |
|
model.gr = 1.0 |
|
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) |
|
model.names = data_dict['names'] |
|
|
|
|
|
labels = np.concatenate(dataset.labels, 0) |
|
c = torch.tensor(labels[:, 0]) |
|
|
|
|
|
plot_labels(labels) |
|
tb_writer.add_histogram('classes', c, 0) |
|
|
|
|
|
check_best_possible_recall(dataset, anchors=model.model[-1].anchor_grid, thr=hyp['anchor_t']) |
|
|
|
|
|
ema = torch_utils.ModelEMA(model) |
|
|
|
|
|
t0 = time.time() |
|
nb = len(dataloader) |
|
n_burn = max(3 * nb, 1e3) |
|
maps = np.zeros(nc) |
|
results = (0, 0, 0, 0, 0, 0, 0) |
|
print('Image sizes %g train, %g test' % (imgsz, imgsz_test)) |
|
print('Using %g dataloader workers' % nw) |
|
print('Starting training for %g epochs...' % epochs) |
|
|
|
for epoch in range(start_epoch, epochs): |
|
model.train() |
|
|
|
|
|
if dataset.image_weights: |
|
w = model.class_weights.cpu().numpy() * (1 - maps) ** 2 |
|
image_weights = labels_to_image_weights(dataset.labels, nc=nc, class_weights=w) |
|
dataset.indices = random.choices(range(dataset.n), weights=image_weights, k=dataset.n) |
|
|
|
mloss = torch.zeros(4, device=device) |
|
print(('\n' + '%10s' * 8) % ('Epoch', 'gpu_mem', 'GIoU', 'obj', 'cls', 'total', 'targets', 'img_size')) |
|
pbar = tqdm(enumerate(dataloader), total=nb) |
|
for i, (imgs, targets, paths, _) in pbar: |
|
ni = i + nb * epoch |
|
imgs = imgs.to(device).float() / 255.0 |
|
|
|
|
|
if ni <= n_burn: |
|
xi = [0, n_burn] |
|
|
|
accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round()) |
|
for j, x in enumerate(optimizer.param_groups): |
|
|
|
x['lr'] = np.interp(ni, xi, [0.1 if j == 2 else 0.0, x['initial_lr'] * lf(epoch)]) |
|
if 'momentum' in x: |
|
x['momentum'] = np.interp(ni, xi, [0.9, hyp['momentum']]) |
|
|
|
|
|
if opt.multi_scale: |
|
sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs |
|
sf = sz / max(imgs.shape[2:]) |
|
if sf != 1: |
|
ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]] |
|
imgs = F.interpolate(imgs, size=ns, mode='bilinear', align_corners=False) |
|
|
|
|
|
pred = model(imgs) |
|
|
|
|
|
loss, loss_items = compute_loss(pred, targets.to(device), model) |
|
if not torch.isfinite(loss): |
|
print('WARNING: non-finite loss, ending training ', loss_items) |
|
return results |
|
|
|
|
|
if mixed_precision: |
|
with amp.scale_loss(loss, optimizer) as scaled_loss: |
|
scaled_loss.backward() |
|
else: |
|
loss.backward() |
|
|
|
|
|
if ni % accumulate == 0: |
|
optimizer.step() |
|
optimizer.zero_grad() |
|
ema.update(model) |
|
|
|
|
|
mloss = (mloss * i + loss_items) / (i + 1) |
|
mem = '%.3gG' % (torch.cuda.memory_cached() / 1E9 if torch.cuda.is_available() else 0) |
|
s = ('%10s' * 2 + '%10.4g' * 6) % ( |
|
'%g/%g' % (epoch, epochs - 1), mem, *mloss, targets.shape[0], imgs.shape[-1]) |
|
pbar.set_description(s) |
|
|
|
|
|
if ni < 3: |
|
f = 'train_batch%g.jpg' % i |
|
res = plot_images(images=imgs, targets=targets, paths=paths, fname=f) |
|
if tb_writer: |
|
tb_writer.add_image(f, res, dataformats='HWC', global_step=epoch) |
|
|
|
|
|
|
|
|
|
|
|
scheduler.step() |
|
|
|
|
|
ema.update_attr(model) |
|
final_epoch = epoch + 1 == epochs |
|
if not opt.notest or final_epoch: |
|
results, maps, times = test.test(opt.data, |
|
batch_size=batch_size, |
|
imgsz=imgsz_test, |
|
save_json=final_epoch and opt.data.endswith(os.sep + 'coco.yaml'), |
|
model=ema.ema, |
|
single_cls=opt.single_cls, |
|
dataloader=testloader, |
|
fast=ni < n_burn) |
|
|
|
|
|
with open(results_file, 'a') as f: |
|
f.write(s + '%10.4g' * 7 % results + '\n') |
|
if len(opt.name) and opt.bucket: |
|
os.system('gsutil cp results.txt gs://%s/results/results%s.txt' % (opt.bucket, opt.name)) |
|
|
|
|
|
if tb_writer: |
|
tags = ['train/giou_loss', 'train/obj_loss', 'train/cls_loss', |
|
'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/F1', |
|
'val/giou_loss', 'val/obj_loss', 'val/cls_loss'] |
|
for x, tag in zip(list(mloss[:-1]) + list(results), tags): |
|
tb_writer.add_scalar(tag, x, epoch) |
|
|
|
|
|
fi = fitness(np.array(results).reshape(1, -1)) |
|
if fi > best_fitness: |
|
best_fitness = fi |
|
|
|
|
|
save = (not opt.nosave) or (final_epoch and not opt.evolve) |
|
if save: |
|
with open(results_file, 'r') as f: |
|
ckpt = {'epoch': epoch, |
|
'best_fitness': best_fitness, |
|
'training_results': f.read(), |
|
'model': ema.ema.module if hasattr(model, 'module') else ema.ema, |
|
'optimizer': None if final_epoch else optimizer.state_dict()} |
|
|
|
|
|
torch.save(ckpt, last) |
|
if (best_fitness == fi) and not final_epoch: |
|
torch.save(ckpt, best) |
|
del ckpt |
|
|
|
|
|
|
|
|
|
n = opt.name |
|
if len(n): |
|
n = '_' + n if not n.isnumeric() else n |
|
fresults, flast, fbest = 'results%s.txt' % n, wdir + 'last%s.pt' % n, wdir + 'best%s.pt' % n |
|
for f1, f2 in zip([wdir + 'last.pt', wdir + 'best.pt', 'results.txt'], [flast, fbest, fresults]): |
|
if os.path.exists(f1): |
|
os.rename(f1, f2) |
|
ispt = f2.endswith('.pt') |
|
strip_optimizer(f2) if ispt else None |
|
os.system('gsutil cp %s gs://%s/weights' % (f2, opt.bucket)) if opt.bucket and ispt else None |
|
|
|
if not opt.evolve: |
|
plot_results() |
|
print('%g epochs completed in %.3f hours.\n' % (epoch - start_epoch + 1, (time.time() - t0) / 3600)) |
|
dist.destroy_process_group() if torch.cuda.device_count() > 1 else None |
|
torch.cuda.empty_cache() |
|
return results |
|
|
|
|
|
if __name__ == '__main__': |
|
parser = argparse.ArgumentParser() |
|
parser.add_argument('--epochs', type=int, default=300) |
|
parser.add_argument('--batch-size', type=int, default=16) |
|
parser.add_argument('--cfg', type=str, default='models/yolov5s.yaml', help='*.cfg path') |
|
parser.add_argument('--data', type=str, default='data/coco128.yaml', help='*.data path') |
|
parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='train,test sizes') |
|
parser.add_argument('--rect', action='store_true', help='rectangular training') |
|
parser.add_argument('--resume', action='store_true', help='resume training from last.pt') |
|
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint') |
|
parser.add_argument('--notest', action='store_true', help='only test final epoch') |
|
parser.add_argument('--evolve', action='store_true', help='evolve hyperparameters') |
|
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket') |
|
parser.add_argument('--cache-images', action='store_true', help='cache images for faster training') |
|
parser.add_argument('--weights', type=str, default='', help='initial weights path') |
|
parser.add_argument('--name', default='', help='renames results.txt to results_name.txt if supplied') |
|
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu') |
|
parser.add_argument('--adam', action='store_true', help='use adam optimizer') |
|
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%') |
|
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset') |
|
opt = parser.parse_args() |
|
opt.weights = last if opt.resume else opt.weights |
|
opt.cfg = check_file(opt.cfg) |
|
opt.data = check_file(opt.data) |
|
print(opt) |
|
opt.img_size.extend([opt.img_size[-1]] * (2 - len(opt.img_size))) |
|
device = torch_utils.select_device(opt.device, apex=mixed_precision, batch_size=opt.batch_size) |
|
|
|
if device.type == 'cpu': |
|
mixed_precision = False |
|
|
|
|
|
if not opt.evolve: |
|
tb_writer = SummaryWriter(comment=opt.name) |
|
print('Start Tensorboard with "tensorboard --logdir=runs", view at http://localhost:6006/') |
|
train(hyp) |
|
|
|
|
|
else: |
|
tb_writer = None |
|
opt.notest, opt.nosave = True, True |
|
if opt.bucket: |
|
os.system('gsutil cp gs://%s/evolve.txt .' % opt.bucket) |
|
|
|
for _ in range(10): |
|
if os.path.exists('evolve.txt'): |
|
|
|
parent = 'single' |
|
x = np.loadtxt('evolve.txt', ndmin=2) |
|
n = min(5, len(x)) |
|
x = x[np.argsort(-fitness(x))][:n] |
|
w = fitness(x) - fitness(x).min() |
|
if parent == 'single' or len(x) == 1: |
|
|
|
x = x[random.choices(range(n), weights=w)[0]] |
|
elif parent == 'weighted': |
|
x = (x * w.reshape(n, 1)).sum(0) / w.sum() |
|
|
|
|
|
mp, s = 0.9, 0.2 |
|
npr = np.random |
|
npr.seed(int(time.time())) |
|
g = np.array([1, 1, 1, 1, 1, 1, 1, 0, .1, 1, 0, 1, 1, 1, 1, 1, 1, 1]) |
|
ng = len(g) |
|
v = np.ones(ng) |
|
while all(v == 1): |
|
v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0) |
|
for i, k in enumerate(hyp.keys()): |
|
hyp[k] = x[i + 7] * v[i] |
|
|
|
|
|
keys = ['lr0', 'iou_t', 'momentum', 'weight_decay', 'hsv_s', 'hsv_v', 'translate', 'scale', 'fl_gamma'] |
|
limits = [(1e-5, 1e-2), (0.00, 0.70), (0.60, 0.98), (0, 0.001), (0, .9), (0, .9), (0, .9), (0, .9), (0, 3)] |
|
for k, v in zip(keys, limits): |
|
hyp[k] = np.clip(hyp[k], v[0], v[1]) |
|
|
|
|
|
results = train(hyp.copy()) |
|
|
|
|
|
print_mutation(hyp, results, opt.bucket) |
|
|
|
|
|
|
|
|