Video-based Pooling and Epistemic Scoring in MONAILabel with CVAT

[bilalcodehub]

Hi MONAILabel Team,

I’m working on a video annotation project using MONAILabel integrated with CVAT. My dataset consists mainly of video clips, each around 30 seconds long at 30 frames per second.

I have a few questions regarding support for video workflows in MONAILabel:

1. Is there a way to provide MONAILabel with a pool of video files directly, rather than pre-extracted images?
2. Can MONAILabel compute epistemic uncertainty or scoring on a per-video basis, instead of just individual frames?
3. Is it possible for MONAILabel to select a video based on its epistemic score for active learning, and then push the full set of extracted frames—preserving temporal order—into a single CVAT task? This would help leverage CVAT’s tracking and interpolation capabilities effectively.

Any guidance on whether the current version supports this workflow—or pointers to relevant documentation—would be greatly appreciated.

Thank you for your support!

Kind regards,
Bilal