VideoPipe

Why we included this project

VideoPipe is a C++ framework for video analysis built as a chain of independent nodes, closer in spirit to NVIDIA's DeepStream but open source and portable across vendors. It runs on NVIDIA, Jetson, Rockchip, Ascend, or plain CPU boxes, and ingests RTSP, RTMP, UDP, or local streams. Nodes handle detection, classification, tracking, feature extraction, and multimodal-LLM inference; structured results can go to Kafka or other sinks and be overlaid on the frame. You bring your own models and pick the inference backend, from OpenCV DNN to TensorRT, Paddle, or ONNX Runtime, so the framework stays compact and leaves hardware and model choices to you. That makes it a practical base for traffic-incident alerts, license-plate recognition, face recognition, or behavior analysis without locking you into one vendor's ecosystem.