Skip to content

Support video_maxlen_ttl for multi-video cases to better control max framelen per video, and divide max_frames_per_video by video duration automatically #7693

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
1 task done
Luffy-ZY-Wang opened this issue Apr 12, 2025 · 0 comments · May be fixed by #7810
Labels
enhancement New feature or request pending This problem is yet to be addressed

Comments

@Luffy-ZY-Wang
Copy link

Luffy-ZY-Wang commented Apr 12, 2025

Reminder

  • I have read the above rules and searched the existing issues.

Description

Multi-video cases need a param to better control video_maxlen_per_video in a refined way so that we can balance the number of frames occupied by each video.
However we currently only have one param video_maxlen which sets a max number of frames globally for every video. And if we input massive multiple videos, it would just truncate some videos behind instead of trying to decrease the sampled frames of previous videos.

Thus we need a new param like video_maxlen_ttl to help set a total max number of sampled frames, and then divide it by durations of videos automatically to balance frames of each video. While video_maxlen remains its function.

To disable video_maxlen_ttl can be as simple as setting video_maxlen_ttl=video_maxlen*max_num_of_videos.

Pull Request

#7700

@Luffy-ZY-Wang Luffy-ZY-Wang added enhancement New feature or request pending This problem is yet to be addressed labels Apr 12, 2025
@Luffy-ZY-Wang Luffy-ZY-Wang linked a pull request Apr 22, 2025 that will close this issue
2 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request pending This problem is yet to be addressed
Projects
None yet
1 participant