However, if you increase the resolution of the input video delivered by RTSP, the delay of the output video may or may not increase over time.
ex1): Delay does not increase for resolution 640x480
ex2): For a resolution of 1920x1280, the delay increases over time.
I did two tests with “video-viewer.py”.
The result was the same.
Since the result is the same for “video-viewer.py”, the processing speed of “videoSource.Capture ()” of “video-viewer.py” is slower than the frame rate delivered by rtsp, and the frame is “videoSource.Capture ()”. I think it’s in the buffer, what about?
If you are running this as 4x independent posenet.py processes, if you want to keep 1920x1080 resolution you may want to look to using DeepStream which is more optimized for handling multiple streams. As you have found, you could also try reducing the RTSP resolution, which often is okay since the DNN models run at a low resolution anyways (IIRC the pose estimation models are 300x300)