Possible mem leak when using queues and consumer thread doesn't keep up with decoding thread.

[NevermindNilas]

Most of my discussion can be seen here:

python/cpython#120610

[gcanat]

Since this library seems unmaintained, I went through the hassle of writing something similar (except it is based on Rust instead of C++). If you want to give it a try: https://github.com/gcanat/video_reader-rs

It has the get_batch method similar to decord, so if you want to get certain frames from a video you can do:

import video_reader
frames = video_reader.get_batch("video.mp4", frame_indices)

Where frame_indices is the list of indices you want. If you only want one frame with index 42 you could do frame_indices=[42]

Additionally, if you want to decode the whole video:

frames = video_reader.decode("video.mp4")

If you want to only get 1 frames out of 4 in the entire video you can do:

frames = video_reader.decode("video.mp4", compression_factor=0.25)

NOTE: regarding your code snippet to benchmark decord you dont even need opencv to get the frame count, once your instantiate the VideoReader object you can do len(vr) to get it. Also, I'm not sure what you are trying to do with a for loop iterating on each frame index. That's not how this library is supposed to be used. You should use the get_batch method instead and get all the frames you want at once.
Anyway, it is true that decord has some memory management issues and that's also why I made video_reader-rs

[NevermindNilas]

Hello,

Thanks for sharing your project with me.
I will check it out and bring some of my questions to your repo in this case