MMVet mmvet Leaderboard
Auto-discovered from papers reporting MMVet (Score). · Metric: Score (higher is better) · 🟢 Updated 8h ago
| # | Model | Score | Paper |
|---|---|---|---|
| 1 | Full framework (Ours) | 68.70 | link |
| 2 | LoRA (default) | 68.70 | link |
| 3 | w/o QA fidelity scoring | 68.40 | link |
| 4 | w/o cycle-consistent captioning | 68.20 | link |
| 5 | w/o prompt perturbation (temp. sampling) | 68.00 | link |
| 6 | QLoRA (4-bit) | 67.90 | link |
| 7 | w/o self-consistency (STE only) | 67.50 | link |
| 8 | SFT (self-generated data) | 67.20 | link |
| 9 | w/o STE (self-consistency only) | 67.10 | link |
| 10 | BLIP3o-8B (base) | 66.60 | link |
| 11 | Full fine-tuning | 66.00 | link |
| 12 | DualToken (sem.+pcpt.) | 37.80 | link |
| 13 | siglip-so400-patch14-384 | 36.00 | link |
| 14 | VILA-U | 25.90 | link |