Is it possible to use process_video with multiple workers?

[zhmiao]

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Question

Hello. Thank you for the great repo. I wonder if there is any way to make the process_video more optimized with parallelism using multiple workers or threading? Or are there any resources I can dig into for this purpose? Thank you very much!

Additional

No response

[github-actions[bot]]

Hello there, thank you for opening an Issue ! 🙏🏻 The team was notified and they will get back to you asap.

[SkalskiP]

Hi @zhmiao, but do you want to read multiple files at once or read the same file faster?

[zhmiao]

Hello @SkalskiP, thanks for responding. I am looking for faster single video processing, as currently it looks like a single for loop over all the frames. I wonder if there are any ways to process the video with parallelism.

[SkalskiP]

Hi, @zhmiao 👋🏻! Let me convert this issue into a discussion and move it into the Q&A section.

[SkalskiP]

@zhmiao, in theory you could try to do something like this:

generators = [
    sv.get_video_frames_generator(source_path='walking.mp4', end=100),
    sv.get_video_frames_generator(source_path='walking.mp4', start=100, end=200),
    sv.get_video_frames_generator(source_path='walking.mp4', start=200, end=300),
    sv.get_video_frames_generator(source_path='walking.mp4', start=300)
]

import threading

def frame_reader(generator_id, frame_generator):
    for frame_id, frame in enumerate(frame_generator):
        print(f"Generator {generator_id}, Frame {frame_id}")

# Create and start threads
threads = []
for i, gen in enumerate(generators):
    t = threading.Thread(target=frame_reader, args=(i, gen))
    t.start()
    threads.append(t)

# Wait for all threads to complete
for t in threads:
    t.join()

But I'm not even sure you will notice any speed improvements.

[zhmiao]

@SkalskiP Thank you very much. This actually makes sense, but as you said, the speed improvement can be limited.