Image Preprocessing

[saitarslanboun]

Hi!

I unfortunately could not find it anywhere, and forgive me if I have missed it.

For example, if I would like to feed images through pretrained mobilenetv3 as follows;

import timm

m = timm.create_model('mobilenetv3_large_100', pretrained=True)
m.eval()
for image in batch:
     image = preprocess(image)
     out = m(image)

do we have to normalize or do any kind of preprocessing before feeding images, or it is already done automatically?

If we have to normalize, do you have any API for that? In case there is or not, can you show me how we can preprocess images before feeding?

Thanks,

[rwightman]

• https://github.com/rwightman/pytorch-image-models/blob/master/validate.py#L180
• https://github.com/rwightman/pytorch-image-models/blob/f8463b8fa9c0490db093b36acfce71fa2363b8c3/timm/data/loader.py#L162
• https://github.com/rwightman/pytorch-image-models/blob/f8463b8fa9c0490db093b36acfce71fa2363b8c3/timm/data/transforms_factory.py#L228
• https://github.com/rwightman/pytorch-image-models/blob/f8463b8fa9c0490db093b36acfce71fa2363b8c3/timm/data/transforms_factory.py#L130