Details about pretrained models in torchvision

Hello,

Are the pretrained weights for the provided CNN models extracted from a state-of-the-art checkpoint? Were they trained with PyTorch or converted from some other implementations in torch/caffe, etc? I am basically wondering whether the accuracy of these models are on-par with their original papers more or less.

Thanks.

Hi, https://github.com/pytorch/vision/issues/260 says they were trained with PyTorch.

1 Like