Visual Question Answering on PathVQA (test)
92.7Overall AccuracyVILA-M3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VILA-M3Number of Parameters=40B2024.11 | 92.7 | — | — | — | |
| Task Spfc. SOTA2024.11 | 91.7 | — | — | — | |
| VILA-M3Number of Parameters=8B2024.11 | 91 | — | — | — | |
| VILA-M3Number of Parameters=13B2024.11 | 91 | — | — | — | |
| VILA-M3Number of Parameters=3B2024.11 | 87.9 | — | — | — | |
| Med-GeminiNumber of Parameters=1.5T2024.11 | 83.3 | — | — | — | |
| Quilt-LLaVAParams=7B, Samples=107K2026.01 | 58.7 | — | — | 15.3 | |
| SFTPathParams=8B, Samples=13K2026.01 | 57 | — | — | 13.3 | |
| LLaVA-MedParams=7B, Samples=60K2026.01 | 56.2 | — | — | 12 | |
| ReinPathParams=8B, Samples=22K2026.01 | 54.4 | — | — | 14.3 | |
| LLaVA-1.5Params=13B, Samples=665K2026.01 | 54 | — | — | 11.7 | |
| MMQAttention Method=BAN, refinement steps (m)=5, meta-models (n)=32021.05 | 48.8 | 13.4 | 84 | — | |
| MMQAttention Method=SAN, refinement steps (m)=5, meta-models (n)=32021.05 | 47.1 | 11.2 | 82.7 | — | |
| MEVFAttention Method=BAN2021.05 | 44.8 | 8.1 | 81.4 | — | |
| MEVFAttention Method=SAN2021.05 | 43.6 | 6 | 81 | — | |
| MAMLAttention Method=BAN2021.05 | 42.9 | 5.9 | 79.5 | — | |
| MAMLAttention Method=SAN2021.05 | 40.5 | 5.4 | 75.3 | — | |
| BIANAttention Method=BAN, ImageNet pre-trained=true2021.05 | 35.6 | 2.9 | 68.2 | — | |
| StANAttention Method=SAN, ImageNet pre-trained=true2021.05 | 30.5 | 1.6 | 59.4 | — |