Could you clerify mean/std preprocessing for EfficienNet

[sh1ng]

Hi,

just noticed that some networks use
mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD

while the another
mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)

Could you elaborate more how prepare image tensors, especially for the second case?

[rwightman]

If you look up those variables, no difference, ends up being less characters not using the convenience variable so it's not always used