I used python 3.7.11 and torchvision 0.2.1, and the visualization results look like the model learns nothing. I was told that different versions of torchvision might lead to different flow images. Could you tell me which version of torchvision and pytorch did you utilize?