Nicolas Hug

Does this overlap with https://github.com/pytorch/vision/pull/8368 ?

Thanks for the feature request @kevinMEH . That sounds reasonable but before we commit to adding a `ToRGB()` transform, have you considered decoding the images directly into RGB format? E.g....

Glad it works. Calling `.convert("RGB")` is actually what happens by default [in the datasets](https://github.com/pytorch/vision/blob/f52fd335000fa4d91371e8505749cd7ce9249cf8/torchvision/datasets/folder.py#L244-L248), but it's true that it's a bit hidden. We could add a new `ToRGB()` transform to...

Thanks for the feature request @davidpicard , yes we should try to support a `mode` parameter (or similar) like in [timm](https://github.com/huggingface/pytorch-image-models/blob/f2fdd97e9f859285363c05988820c9350b737e59/timm/data/mixup.py#L99)

Thank you for the report @DavidFM43 . Unless I'm missing something, the way things are done in `to_tensor()` is just plain wrong (and the error you get in `pil_to_tensor` is...

Thanks for opening the issue @atharvas . As @pmeier suggested, we'd be happy to look at a PR. Did you find that `masks_to_boxes` was a bottleneck for you? Is it...

Thanks for the PR @hmaarrfk . We don't support 3.12 yet but when the time comes, we'll be happy to re-consider this PR. I genuinely hope that this is *not*...

Thank you for the ping @hmaarrfk . We'll merge the PR in due time, when 3.12 support becomes official for Pytorch and torchvision. I'm sure you're experienced enough to know...

Failures are unrelated, let's merge

I recently re-worked our extension compilation logic and the `image.so` extension is now always built, unconditionally. It doesn't depend anymore on the existence of libjpeg or libpng (because we vendor...