VideoFrame2Tensor provides an efficient pipeline for importing video frames directly into machine learning tensors, eliminating CPU-GPU data copies and big buffer transfers. It leverages hardware-accelerated preprocessing using D3D12 APIs and DirectML operators for real-time ML applications on Windows.
explainer.md— Technical explainer and API usage details.example/semantic_segmentation/— Example implementation for real-time video semantic segmentation, it's a subset of WebNN Samples