|
|
|
@ -133,9 +133,13 @@ def train(hyp):
|
|
|
|
|
with open(results_file, 'w') as file:
|
|
|
|
|
file.write(ckpt['training_results']) # write results.txt
|
|
|
|
|
|
|
|
|
|
# epochs
|
|
|
|
|
start_epoch = ckpt['epoch'] + 1
|
|
|
|
|
assert opt.epochs > start_epoch, '%s has already trained %g epochs. --epochs must be greater than %g' % \
|
|
|
|
|
(opt.weights, ckpt['epoch'], ckpt['epoch'])
|
|
|
|
|
if epochs < start_epoch:
|
|
|
|
|
print('%s has been trained for %g epochs. Fine-tuning for %g additional epochs.' %
|
|
|
|
|
(opt.weights, ckpt['epoch'], epochs))
|
|
|
|
|
epochs += ckpt['epoch'] # finetune additional epochs
|
|
|
|
|
|
|
|
|
|
del ckpt
|
|
|
|
|
|
|
|
|
|
# Mixed precision training https://github.com/NVIDIA/apex
|
|
|
|
@ -166,7 +170,7 @@ def train(hyp):
|
|
|
|
|
|
|
|
|
|
# Testloader
|
|
|
|
|
testloader = create_dataloader(test_path, imgsz_test, batch_size, gs, opt,
|
|
|
|
|
hyp=hyp, augment=False, cache=opt.cache_images, rect=True)[0]
|
|
|
|
|
hyp=hyp, augment=False, cache=opt.cache_images, rect=True)[0]
|
|
|
|
|
|
|
|
|
|
# Model parameters
|
|
|
|
|
hyp['cls'] *= nc / 80. # scale coco-tuned hyp['cls'] to current dataset
|
|
|
|
|