Caption Retrieval on M2M
17.3HIT@1SigLIP-Q
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SigLIP-Q2026.05 | 17.3 | 26 | 38.4 | 52.7 | 67.9 | 80.5 | |
| ToolMerge + GRPOPlanner=Qwen3-VL-8B, Optimization Strategy=GRPO2026.05 | 15.9 | 33.6 | 48.8 | 61.5 | 75.1 | 85.6 | |
| ToolMergePlanner=Qwen3-VL-8B, Mode=zero-shot2026.05 | 14.7 | 29.4 | 43.4 | 58.1 | 73.6 | 85.3 | |
| WFSVenue=CVPR 262026.05 | 11.4 | 18.6 | 27 | 35.7 | 42.4 | 52.6 | |
| Uniform2026.05 | 4 | 5.3 | 11.5 | 18.3 | 37.6 | 61.7 | |
| LIFVenue=NeurIPS 252026.05 | 4 | 7 | 13 | 20.1 | 27.4 | 36.5 | |
| BOLTVenue=CVPR 252026.05 | 1.1 | 4.4 | 15.5 | 29.1 | 47.8 | 71.9 | |
| AKSVenue=CVPR 252026.05 | 0 | 21.7 | 27.9 | 36.4 | 50.7 | 65.9 |