Multi-image Understanding on MIRB (test)
60.8AccuracyQwen2VL-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2VL-7B2026.01 | 60.8 | |
| InternVL2-Llama3-76B2026.01 | 58.2 | |
| GPT-4V2026.01 | 53.1 | |
| attention-masking strategybackbone=LLaVA-OV-7B, attention masking=true2026.01 | 51 | |
| InternVL2-8B2026.01 | 48.6 | |
| LLaVA-OV-7B2026.01 | 47.2 | |
| Qwen2VL-2B2026.01 | 45.9 | |
| procedural data-generation strategybackbone=LLaVA-OV-0.5B, attention masking=false2026.01 | 32.8 | |
| LLaVA-OV-0.5B2026.01 | 31.8 | |
| LLaVA-v1.5-7B2026.01 | 28.5 | |
| attention-masking strategybackbone=LLaVA-OV-0.5B, attention masking=true2026.01 | 28.5 | |
| InternVL2-2B2026.01 | 25 |