Visual Attribution on Thinking-Model Attribution Dataset Overall
72Avg. LDSvStream
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| vStreamModel=Qwen3-VL-8B-Thinking2026.04 | 72 | 1 | 0.024 | |
| AttnLRPModel=Qwen3-VL-8B-Thinking2026.04 | 71 | 0.89 | 2.6 | |
| AttnLRPModel=Cosmos-Reason1-7B2026.04 | 71 | 0.89 | 2.55 | |
| vStreamModel=GLM-4.1V-9B-Thinking2026.04 | 70 | 0.98 | 0.026 | |
| AttnLRPModel=MiMo-VL-7B2026.04 | 70 | 0.91 | 2.34 | |
| vStreamModel=MiMo-VL-7B2026.04 | 70 | 1.02 | 0.022 | |
| vStreamModel=Cosmos-Reason1-7B2026.04 | 70 | 1 | 0.024 | |
| TAMModel=Qwen3-VL-8B-Thinking2026.04 | 69 | 0.99 | 1.9 | |
| AttnLRPModel=GLM-4.1V-9B-Thinking2026.04 | 69 | 0.84 | 2.84 | |
| TAMModel=Cosmos-Reason1-7B2026.04 | 69 | 0.96 | 1.86 | |
| TAMModel=GLM-4.1V-9B-Thinking2026.04 | 68 | 0.96 | 2.07 | |
| TAMModel=MiMo-VL-7B2026.04 | 67 | 0.95 | 1.71 | |
| InputGradModel=Qwen3-VL-8B-Thinking2026.04 | 64 | 0.76 | 2.8 | |
| InputGradModel=MiMo-VL-7B2026.04 | 64 | 0.79 | 2.52 | |
| InputGradModel=GLM-4.1V-9B-Thinking2026.04 | 63 | 0.84 | 3.05 | |
| InputGradModel=Cosmos-Reason1-7B2026.04 | 62 | 0.78 | 2.75 | |
| AttentionModel=GLM-4.1V-9B-Thinking2026.04 | 41 | 0.31 | 0.022 | |
| AttentionModel=Cosmos-Reason1-7B2026.04 | 41 | 0.35 | 0.02 | |
| AttentionModel=Qwen3-VL-8B-Thinking2026.04 | 40 | 0.29 | 0.02 | |
| AttentionModel=MiMo-VL-7B2026.04 | 40 | 0.31 | 0.018 | |
| RandomModel=Qwen3-VL-8B-Thinking2026.04 | 31 | 0.1 | 0.01 | |
| RandomModel=GLM-4.1V-9B-Thinking2026.04 | 31 | 0.12 | 0.011 | |
| RandomModel=Cosmos-Reason1-7B2026.04 | 30 | 0.15 | 0.01 | |
| RandomModel=MiMo-VL-7B2026.04 | 29 | 0.14 | 0.009 |