Are the exact training details (training data + hyperparameters) for the various models distributed with Torchvision written down somewhere? In particular, I always assumed that the VGG16 model (vgg16-397923af.pth) that Torchvision can automatically download is just a repackaged version of the pretrained model distributed by the VGG paper authors at their website. That seems not to be the case.
Moreover, what are the licensing terms of the pretrained models distributed with Torchvision?