PyNVVideoCodec provides Python bindings for NVIDIA Video Codec SDK which enabled C++ developers for over a decade with hardware accelerated video encode and decode for both Windows and Linux. Python developers can now easily access NVENC and NVDEC acceleration units on the GPU to encode and decode video data to/from GPU memory. Depending on your GPU, different number and generation of hardware units are available. Visit https://developer.nvidia.com/video-codec-sdk for more details including benchmarks data points.

PyNvVideoCodec is a part of Vision Cloud package, an initiative to power Python developers with reference libraries and tools including CV-CUDA, NVImageCodec and PyNVVideoCodec.

Current PyNVVideoCodec version includes support for:

• Simple APIs for hardware accelerated video decode and encode
• H.264 / HEVC / AV1 Codecs
• NV12, P016, YUV 4:4:4 (8 and 10 bit) Decode Surface formats
• NV12, P010, YUV 4:4:4 (8 and 10 bit), ARGB, ABGR Encode Surface formats
• DLPack and CUDA Array Interface Data exchange formats
• CV-CUDA and PyTorch frameworks
• Python sample applications that demonstrate the usage of APIs
• Pip install script