Launch: Run Vision Models on Multiple Streams

Roboflow Inference, a high-performance computer vision inference server, now supports running models on multiple video streams simultaneously, making it suitable for large-scale deployments such as monitoring multiple CCTV feeds from a single server. The InferencePipeline method allows users to deploy any model hosted on Roboflow and process video streams from RTSP sources, webcams, or video files using a callback function for predictions. Users must first install Roboflow Inference and authenticate with an API key to access models, including over 50,000 public ones available on Roboflow Universe. The pipeline can be customized with different models and callback functions, providing flexibility for various applications, including traffic analytics and commercial deployments. Roboflow offers additional support for enterprise customers and guidance on selecting appropriate cameras and lenses for optimal computer vision project outcomes.