My question is:
Of all the available CNN architectures in fastai (Resnets(18,34,50,101,152), xResnets(18,34,50,101,152), squeezenets, densenets, VGG , alexnet), what are each one of these pretrained on? Are all these pretrained on ImageNet only?
Also, are there going to be more architectures available in fastai v2?
Cheers, Stay Safe
You can use any PyTorch model with the fastai framework. That’s all those models are. They come from torchvision’s model zoo. All are pretrained on ImageNet unless their documentation states otherwise. (XResNet is a fastai exclusive)
My bad. I thought that these models and weights were provided by fastai , and I spent a lot of time searching for the origin of the architectures and the pretrained weights in the source code. The More I know! Thanks a lot