libffcv / ffcv

FFCV: Fast Forward Computer Vision (and other ML workloads!)
https://ffcv.io
Apache License 2.0
2.79k stars 180 forks source link

Equivalent of random crop in ffcv #339

Open meghbhalerao opened 11 months ago

meghbhalerao commented 11 months ago

Dear Authors, I am trying to reproduce a paper, which uses RandomCrop transform of torchvision transforms as here - https://pytorch.org/vision/main/generated/torchvision.transforms.RandomCrop.html#torchvision.transforms.RandomCrop - but I want to speed it up using ffcv, so I want the exact same augmentation in ffcv, but it seems like I can not find an equivalent of random crop in the ffcv decoders and transforms as seen https://docs.ffcv.io/api/transforms.html and https://docs.ffcv.io/api/decoders.html

Am I looking at the right place? please let me know, this is also related to another issue here - https://github.com/libffcv/ffcv/issues/337

thanks for your time and please let me know if I am missing anything!