Caption Token Attribution on COCO
64.05Insertion ScoreGreedy
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GreedyModel=Qwen2.5-VL-3B2026.06 | 64.05 | 43.72 | 59.46 | 28.58 | 69.08 | 69.51 | 4,168.62 | 1.53 | |
| PhaseWinModel=Qwen2.5-VL-3B2026.06 | 63.51 | 45.22 | 57.36 | 30.52 | 67.86 | 68.35 | 1,412.71 | 4.8 | |
| GreedyModel=Qwen2.5-VL-7B2026.06 | 62.84 | 43.5 | 57.21 | 28.15 | 70.81 | 70.64 | 2,931.52 | 2.14 | |
| PhaseWinModel=Qwen2.5-VL-7B2026.06 | 61.55 | 44.67 | 55.35 | 29.68 | 70.15 | 69.95 | 1,401.6 | 4.39 | |
| RISEModel=Qwen2.5-VL-3B2026.06 | 56.08 | 50.87 | 47.71 | 38.93 | 66 | 66.45 | 5,000 | 1.12 | |
| RISEModel=Qwen2.5-VL-7B2026.06 | 55.4 | 50.19 | 46.1 | 37.71 | 68.68 | 68.21 | 5,000 | 1.11 | |
| IGOS++Model=Qwen2.5-VL-3B2026.06 | 53.76 | 52.96 | 43.88 | 42.66 | 65.74 | 66.2 | — | — | |
| GradientModel=Qwen2.5-VL-3B2026.06 | 53.65 | 53.15 | 43.97 | 42.98 | 65.75 | 66.15 | — | — | |
| IGOS++Model=Qwen2.5-VL-7B2026.06 | 53.5 | 52.19 | 43.13 | 40.97 | 68.74 | 68.16 | — | — | |
| LLaVACAMModel=Qwen2.5-VL-7B2026.06 | 53.4 | 53.57 | 43.47 | 43.62 | 68.9 | 68.24 | — | — | |
| GradientModel=Qwen2.5-VL-7B2026.06 | 52.79 | 52.48 | 42.1 | 41.49 | 68.42 | 67.91 | — | — | |
| LLaVACAMModel=Qwen2.5-VL-3B2026.06 | 52.48 | 54.6 | 41.84 | 45.3 | 65.47 | 65.99 | — | — |