Maybe a Bug

[QuarTerll]

In train.py, between line 54 and 69,

ASL/train.py

Lines 54 to 69 in 7068ff0

	val_dataset = CocoDetection(data_path_val,
	instances_path_val,
	transforms.Compose([
	transforms.Resize((args.image_size, args.image_size)),
	transforms.ToTensor(),
	# normalize, # no need, toTensor does normalization
	]))
	train_dataset = CocoDetection(data_path_train,
	instances_path_train,
	transforms.Compose([
	transforms.Resize((args.image_size, args.image_size)),
	CutoutPIL(cutout_factor=0.5),
	RandAugment(),
	transforms.ToTensor(),
	# normalize,
	]))

It says that

#normalize, # no need, toTensor does normalization

I have different thought. To Tensor scales images to (0, 1) while normalization to (-1, 1).
This should be different.

I do not test for this. Please Check it.

Thank you. :)

[QuarTerll]

And another question，

If so， why you guys don't use normalization?

The reason ask this is that my f1-score shows 10% differences at most when I use normalization with your code with ResNet50