Contributing to pytorch and torchvision


I have created to datasets that are not available under the torchvision package. These datasets are commonly used in the computer vision community, yet they are not as used as CIFAR10, CIFAR100… However, I think there are interesting to be part of torchvision package.

The datastes are: and

Is there are any standard to make a pull request?.


1 Like

Thanks !
pinging @smth, he will know about it :slight_smile:

okei, @smth @albanD I am finishing training some deep models pretrained on imagenet from torchvision.model to check everything is ok. After that I will be able to git push. I have followed the standards on the torchvision.datasets. I tried to follow the same guideline. The only thing I miss is some kind of checksum for the downloaded dataset and preprocessing. I have also add some features to the dataset based on numpy operations. I will make that clear in the dataset python class.

Okei, datasets seems to be correct. Getting 90% accuracy on cars and 80% accuracy on birds which are among the best results published for these tasks (at least to my knowledge). I need to make two little changes to the classes and they will be ready to upload.