VideoPipe

VideoPipe is a video analysis framework for integrating and deploying computer vision AI models, including multimodal large models. It is written in native C++ and does not depend on heavy third party frameworks such as GStreamer or GObject. Developers build an application by connecting plugins, which the framework calls nodes, into a pipeline: they define the node types they need, create instances and attach them together. A pipeline can take multiple inputs and outputs, and its branches can be merged or split. It is aimed at engineers building video analytics products for smart devices, edge servers and data centers. Built in plugins cover video ingestion, encoding and decoding, algorithm inference, object tracking, message push, recording and screenshots, and video streaming. Custom inference nodes can use backends such as OpenCV DNN, TensorRT, PaddleInference, ONNX Runtime, RKNN or large model API services. A visual debugging tool shows the live state of a running pipeline, including buffer sizes, frame rate and latency at each stage, to help locate bottlenecks. Deployment is cross platform and adapts to each hardware vendor's SDK without maintaining separate source trees. The site lists example applications including face recognition, license plate recognition cameras, traffic violation detection, face swapping, image search, security monitoring and lane line detection. It also links to video tutorials and the GitHub repository. The site is written in Chinese, and the author offers consulting by messaging. No pricing is published.