Efficientnet B4 pretrained weights and preprocessing

[anticsalle]

Hi there,

I would like to ask for a few details concerning the timm efficientnet b4 - if pretrained model with 1000 clasees is used, is that model trained on imagent ? If so, what classes encoding dictinary was used ? I have tried loading a few dog images for that model and it gives outputs for 618 (3x) and 731 (1x) class which doesen't seem right (link for classes encoding for imagenet I have encountered >> https://gist.github.com/yrevar/942d3a0ac09ec9e5eb3a; >> https://github.com/anishathalye/imagenet-simple-labels/blob/master/imagenet-simple-labels.json)

And I would also like to ask if any specific preprocessing was used for training the model - I assumed minmax normalization would do no harm, so I tried dogs images set to range [0,1]

Looking forward for your response,
Best regards,