Multi-image retrieval on MS-COCO Random Setting N=4 (test)
98.66AccuracyLAD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LADBackbone=LLaVA-OneVision-8B2026.05 | 98.66 | 0.88 | 96.5 | |
| LADBackbone=Qwen2.5-VL-3B2026.05 | 94.94 | 4.2 | 84.3 | |
| LADBackbone=InternVL-3-8B2026.05 | 94.84 | 5.93 | 80.7 | |
| CoTBackbone=InternVL-3-8B2026.05 | 77.96 | 12.45 | 37.7 | |
| PriDeBackbone=InternVL-3-8B2026.05 | 76.32 | 18.33 | 35.9 | |
| Self-ConsistencyBackbone=InternVL-3-8B2026.05 | 69.2 | 24.06 | 22 | |
| VanillaBackbone=InternVL-3-8B2026.05 | 69.04 | 28.38 | 18.5 | |
| PriDeBackbone=LLaVA-OneVision-8B2026.05 | 68 | 33.07 | 32.1 | |
| VanillaBackbone=LLaVA-OneVision-8B2026.05 | 67.52 | 25.43 | 19.9 | |
| InstructionBackbone=InternVL-3-8B2026.05 | 67.52 | 26.27 | 19.6 | |
| PriDeBackbone=Qwen2.5-VL-3B2026.05 | 63.76 | 36.24 | 17.7 | |
| Self-ConsistencyBackbone=Qwen2.5-VL-3B2026.05 | 63.24 | 20.35 | 15.7 | |
| CoTBackbone=Qwen2.5-VL-3B2026.05 | 54.16 | 36.75 | 8.2 | |
| VanillaBackbone=Qwen2.5-VL-3B2026.05 | 51.7 | 36.27 | 6.61 | |
| InstructionBackbone=LLaVA-OneVision-8B2026.05 | 50.5 | 34.61 | 5.9 | |
| InstructionBackbone=Qwen2.5-VL-3B2026.05 | 48.96 | 35.8 | 4.2 | |
| CoTBackbone=LLaVA-OneVision-8B2026.05 | 45.4 | 34.15 | 3.31 | |
| Self-ConsistencyBackbone=LLaVA-OneVision-8B2026.05 | 41.9 | 19.09 | 0 |