A single 22B-parameter model jointly restores degraded video and audio of historical films, beating prior separate restorers on visual, audio, and sync metrics.
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
OmniVR: Joint Video-Audio Conditional Generation for Restoring Degraded Historical Films
A single 22B-parameter model jointly restores degraded video and audio of historical films, beating prior separate restorers on visual, audio, and sync metrics.