TorchVision 0.29 is out! It comes with two major changes: ABI stability, and deprecation of the image decoders and encoders (now in TorchCodec)!
ABI Stability with torch 2.14
TorchVision is now ABI stable w.r.t. torch 2.14! This means that torchvision 0.29 will be compatible with future versions of torch: 2.15, 2.16, etc. You won’t need to install a new version of TorchVision when you upgrade torch.
As a result, we might stop releasing TorchVision in sync with pytorch. But TorchVision is still actively maintained and developed: we’ll still be pushing releases, just not with the same cadence.
Thanks to Adrian Abeyta @adabeyta for the fantastic porting effort!
PRs: #9524, #9597, #9598, #9605, #9612, #9610, #9614, #9584, #9617, #9618, #9619, #9620, #9582, #9573, #9625, #9623, #9626, #9583, #9633, #9572, #9549, #9533, #9535, #9539, #9543, #9550, #9552, #9555, #9557, #9558, #9554
Image decoders and encoders are deprecated. Use TorchCodec!
The image decoders and encoders in torchvision.io are now deprecated, and they will be removed in a future release. They are now available in torchcodec >= 0.16, where they are significantly more capable. You’ll just need to pip install torchcodec, and you can refer to this migration guide for migrating your code (most APIs for decoding are the same).
This finalizes a clear separation of concerns for the three media-processing libraries of PyTorch: torchcodec is for decoding and encoding all media (images, videos, and audio), while torchvision and torchaudio focus on the transforms.
Bug fixes
[ops] Fix for deformable convolution kernels always running on default stream (#9522)
[ops, MPS] Fix gradient overaccumulation in ROI ops (#9563, #9510)
[transforms] Fix JPEG transform for non-contiguous batches (#9615)
Contributors
🎉 We're grateful for our community, which helps us improve Torchvision by submitting issues and PRs, and providing feedback and suggestions. The following persons have contributed patches for this release:
Adrian Abeyta, Andrey Talman, Dmitry Nikolaev, Irakli Salia, Jeff Daily , Kasra Ghodsi, Nicolas Hug, Nikita Shulga, Simon Byrne, Yutao Xu, Zhewen