Preprocessing for each network

[MayStepanyan]

Hi, thanks for the huge work you've done! The package is very helpful

I have an issue figuring out what preprocessing each network's inputs need. Is there a list or some other place where I can look up this info?

[leondgarse]

Ya, I didn't make a list for that info. Currently you can get them by:

""" Get model rescale mode """
from keras_cv_attention_models import efficientnet
mm = efficientnet.EfficientNetV2B0()
print(mm.preprocess_input.rescale_mode)
# torch

""" Get the related mean std value for rescale mdoe """
from keras_cv_attention_models.imagenet import data
mean, std = data.init_mean_std_by_rescale_mode(mm.preprocess_input.rescale_mode)
print(mean, std)
# 0 255.0

Generally, it's just resize an image to mm.input_shape[1:-1], then apply mean and std (image - mean) / std.

[MayStepanyan]

Great, thanks.