We are pleased to announce the release of PyNvVideoCodec 2.2, the latest version of our GPU-accelerated Python library for video decoding, encoding, and transcoding , all with the simplicity and flexibility of Python.
PyNvVideoCodec 2.2 expands platform support to NVIDIA Jetson and NVIDIA DGX Spark, while also introducing the ability to decode video directly from in-memory data.
These new capabilities make it easier to build high-performance video and AI pipelines across robotics, edge, and data centre environments.
New Features in PyNvVideoCodec 2.2
- NVIDIA Jetson support: Brings easy-to-use, GPU-accelerated Python video APIs to Jetson, making it easy to build video and AI pipelines for robotics and other edge devices.
- NVIDIA DGX Spark support: Unlocks high-throughput, Python-based video decode and encode on DGX Spark, accelerating video curation, experimentation, prototyping, training, and deployment of video and AI pipelines.
- Access to in-memory video data: Reads video directly from Python byte data and streams, allowing video input from network buffers or cloud storage (such as S3) without writing to temporary files, while still supporting random frame access and batch retrieval.
- New samples:
- Face protection: Jupyter notebook that demonstrates an important privacy use case on Jetson—detecting faces in video with a TensorRT-optimized RetinaFace model and blurring them using GPU-accelerated Gaussian blur to protect identities in video.
- Video decode from memory: Illustrates decoding video directly from in-memory Python byte data and seekable binary streams (such as bytes, bytearray, memoryview, BytesIO, or file handles).
- Video transcode: Shows end-to-end GPU video transcoding—demuxing, decoding, encoding, and muxing—using hardware-accelerated video decode and encode.
- Video encode with muxing: The encode sample now shows how to write encoded video directly into container formats such as MP4 and MOV.