Region-level detailed image captioning on MDVP-Bench Box Caption
178.6Natural ScoreGAR-8B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GAR-8BSetting=Zero-shot, Judger=GPT-4o, Mask Generation=SAM conditioned on bounding boxes2025.10 | 178.6 | 149.1 | 117.2 | 123 | |
| GAR-1BSetting=Zero-shot, Judger=GPT-4o, Mask Generation=SAM conditioned on bounding boxes2025.10 | 152.6 | 149.6 | 103.7 | 115.3 | |
| Osprey-7BSetting=Zero-shot, Judger=GPT-4o, Mask Generation=SAM conditioned on bounding boxes2025.10 | 107.7 | 99.4 | 70 | 81.3 | |
| DAM-3BSetting=Zero-shot, Judger=GPT-4o, Mask Generation=SAM conditioned on bounding boxes2025.10 | 87 | 127.7 | 79.4 | 76.4 | |
| PAM-3BSetting=Zero-shot, Judger=GPT-4o, Mask Generation=SAM conditioned on bounding boxes2025.10 | 71.4 | 94.3 | 86.8 | 84.5 |