TVM Streamer is our product name for a video processing application called tvminfer implemented in the GStreamer framework using the TVM deep learning compiler stack.
There is a growing demand for video camera surveillance systems in the context of crime prevention and safety around the world. We are building a video streaming infrastructure that can efficiently process a large amount of 4K video in real time, and we are using TVM for the video inference part to realize a high-performance video processing application.
We have compared the performance of TVM Streamer with similar software such as Intel DL Streamer and NVIDIA DeepStream SDK in various scenarios, and have confirmed the performance advantage of TVM Streamer in several use cases.
In this session, we will present the outline, method of implementation, and performance evaluation results of TVM Streamer, and discuss the versatility of TVM in solving real-world problems.