Visual Relational Reasoning in videos on VRR-QA 1.0 (test)
90.07Average AccuracyAdaptive Dense Evidence Refinement
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Adaptive Dense Evidence RefinementStage=Final system, Backbone=Gemini2026.05 | 90.07 | 87.81 |
Loading the SOTA2 catalog…