IsoVRBench: Synthetic Interventions Yield 160K Video Preference Pairs Without Human Labeling
IsoVRBench generates 160K video preference pairs by applying targeted interventions (e.g., shuffling or freezing frames) to anchor videos, automatically yielding fine-grained labels without human annotation and exposing temporal weaknesses in current video reward models.
2026-10-14 ~ 2026-10-14 · 2 related posts
- IsoVRBench: 160K synthetic video pairs via targeted interventions train reward models without human labels — StellaLisy · 2026-10-14
- IsoVRBench: targeted interventions expose temporal consistency failures in video reward models — StellaLisy · 2026-10-14