I am going through someone’s project and noticed that they are downloading resnet models using torch.utils.model_zoo module.
I have only used torchvision.models so far
What is the difference between loading pretrained models via model_zoo vs torchvision.models?
The model_zoo was moved to the torch.hub, if I’m not mistaken.
torch.hub is an easy way to load pretrained models from public GitHub repositories and you can find the currently supported models here.
For a resnet you could either use the torchvision implementation or load these models from the hub, which should use the same checkpoint and architectures.
One advantage of the hub is that you can easily add support for your current repository and share your model easier than trying to add your current model to e.g. torchvision.