This is a great guide! Although the frame skipping method might be the easiest to implement, I rather work with different threads because it allows more flexibility depending on the final machine.

I usually split the application into two threads connected with a buffer:
1 - consumes the frames and displays them into an ui. For each new frame, I check weather the buffer is empty. In case it is empty, I add the current frame.
2 - processing thread. Runs in an infinite loop. Checks if there is a frame in the buffer and process it.

In this way, the application achieves the fastest possible inference time.

Great content!