s3d.pytorch
s3d.pytorch copied to clipboard
Problem about normalization
Hi @qijiezhao , thank you for sharing your implementation. I am wondering if your pre-trained RGB_imagenet.pkl is just converted from deepmind's tf model weights or not. According to I3D's repository, pixels should be scaled between -1 and 1, but I can't find this normalization step in your code, is this a negligence or for other reasons, thanks again.