Multi-Image Fine-Grained Question Answering on MIFG-QA
50.24Accuracy (Nature)IMAgent
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| IMAgentE2E=true, Param=7B2025.12 | 50.24 | 48.73 | 49.21 | |
| GPT-4o2025.12 | 49.26 | 47.55 | 48.11 | |
| Pixel ReasonerE2E=false, Param=7B2025.12 | 41.87 | 45.72 | 44.49 | |
| DeepeyesE2E=true, Param=7B2025.12 | 38.75 | 43.84 | 38.36 | |
| LLaVA-OneVisionParam=7B2025.12 | 37.93 | 39.49 | 38.99 | |
| Qwen2.5-VLParam=7B2025.12 | 36.94 | 44.34 | 41.98 | |
| Qwen2.5-VL-CoTParam=7B2025.12 | 36.35 | 40.65 | 39.31 |