NVIDIA DeepStream — видеоаналитика и пайплайны

★ 7.6 · content

deepstream-dev is a Claude Code skill that assists in building video analytics pipelines with the NVIDIA DeepStream SDK using the Python pyservicemaker API. It provides a structured reference for GStreamer pipeline architecture covering sources (nvurisrcbin, nvmultiurisrcbin), stream muxing with nvstreammux, TensorRT inference via nvinfer and nvinferserver, multi-object tracking with nvtracker, OSD overlays, and rendering. The skill encodes critical rules around NVMM zero-copy GPU buffers, metadata iterators, nvinfer config formats (YAML and INI), tee-based pipeline splits, and Kafka integration via nvmsgbroker. Targets Ubuntu 24.04 on x86_64 and ARM64/Jetson with NVIDIA Driver 590+, CUDA 13.1, and TensorRT 10.14. Ideal for GPU-accelerated computer vision tasks such as object detection and tracking from RTSP streams, cameras, or video files, including Ultralytics YOLO models across generations.