Multimodal Retrieval-Augmented Generation on CRAG-MM
20.5TruthfulnessPixSearch-13B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=Full2026.01 | 20.5 | 45.9 | 18.7 | 25.4 | |
| Llama-3.2-11BBackbone=Llama-3.2-11B-Vision, Search Strategy=Whole Image2026.01 | 11.5 | 46.6 | 18.3 | 35.1 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=Whole Image2026.01 | 5.1 | 44.8 | 15.5 | 39.7 | |
| Llama-3.2-11BBackbone=Llama-3.2-11B-Vision, Search Strategy=No Search2026.01 | 0.1 | 34.2 | 31.7 | 34.1 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=Full2026.01 | -3 | 37.7 | 21.6 | 40.7 | |
| Llama-3.2-11BBackbone=Llama-3.2-11B-Vision, Search Strategy=Whole Image2026.01 | -8.6 | 35.3 | 20.3 | 43.9 | |
| PLUM-13BBackbone=LLaVA-13B2026.01 | -10.1 | 39.4 | 11 | 49.5 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=Text Query2026.01 | -14.8 | 34.4 | 16.4 | 49.2 | |
| Llama-3.2-11BBackbone=Llama-3.2-11B-Vision, Search Strategy=No Search2026.01 | -16.9 | 24.4 | 34.4 | 41.3 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=No Search2026.01 | -18.8 | 31.7 | 17.8 | 50.5 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=Whole Image2026.01 | -19 | 31.5 | 18.1 | 50.5 | |
| GroundedSAM+LLaVA-13BBackbone=LLaVA-13B, Search Strategy=Pipeline Method (Region-Level)2026.01 | -23.6 | 25.7 | 25 | 49.3 | |
| GroundedSAM+LLaVA-13BBackbone=LLaVA-13B, Search Strategy=Pipeline Method (Region-Level)2026.01 | -30.4 | 23.4 | 22.8 | 53.8 | |
| PLUM-13BBackbone=LLaVA-13B2026.01 | -35.2 | 25.3 | 14.5 | 60.5 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=No Search2026.01 | -40.4 | 19.8 | 20 | 60.2 | |
| PixSearch-13BBackbone=LLaVA-13B, Search Strategy=Text Query2026.01 | -44.5 | 18.2 | 19.1 | 62.7 | |
| LLaVA-13BBackbone=LLaVA-13B, Search Strategy=None2026.01 | -59.1 | 16.1 | 8.7 | 75.2 | |
| LLaVA-13BBackbone=LLaVA-13B, Search Strategy=None2026.01 | -69 | 10.3 | 10.4 | 79.3 |