Question about image normalization used during training

[smaz30]

Hi,

First of all, thank you for sharing this work and for making the code available.

I have a quick question regarding the image preprocessing used during training and inference. Could you clarify how the input images are normalized before being fed to the network?

Specifically, are the images:

• Scaled to the ([0, 1]) range only, or
• Normalized using ImageNet statistics (mean = [0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225]), or
• Processed using a different normalization scheme?

I couldn't find this detail in the documentation, and I would like to make sure I reproduce the setup correctly.

Thank you in advance for your help.