Does ml.net add a pipeline similar to nvidia deepstream?

In real-time streaming image analysis, such as rtmp and rtsp, we use emgucv, opencvsharp... or other tools to pull the stream, and capture the image in it for object detection and object classification. Video stream -> GPU decoding -> CPU conversion image format -> ML.net GPU detection, and then returns the result -> C# statistics and drawing the box.

This process seems inefficient and the CPU usage is very high. If ML.net's pipelines can pull images -> detection -> drawing are all done on the GPU, it should greatly reduce the load.