TorchVision
summary
TorchVision now lets teams export complete pipelines to C++: TorchScript and ONNX export for all pre-trained models and mixed-precision inference work out of the box. Transforms are actual nn.Modules that accept Tensors on GPU and can be exported, enabling GPU-accelerated transforms with Tensor input. Image IO adds native JPEG/PNG read/write and a raw-bytes uint8 tensor path, i.e. native JPEG/PNG image IO and raw byte tensors. Video gets a beta frame-based reader with audio and an iterator API.
AI-written summary · about
chapters (3)
related talks
11mWhat's New in Torch.onnx
05mPyTorch ONNX Export Support
09mTorchvision 0.4 with Support for Video: A Quick Introduction
10mLightning Talk: Streamlining Model Export with the New ONNX Exporter
05mWhat's New in the Torch Ecosystem
30mThe Evolution of ONNX: Driving Interoperability in AI...
10mResearch to Production: PyTorch JIT/TorchScript Updates
09mTogether Goes Brrr: Threading Research & Production with Torch Compile