Visual Question Answering on M3CoT
61.2AccuracyAIM-CoT
Evaluation Results
| Method | Links | |
|---|---|---|
| AIM-CoTBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 61.2 | |
| ICoTBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 59.1 | |
| AIM-CoTBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 58.7 | |
| DDCoTBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 57.5 | |
| CCoTBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 56.9 | |
| ICoTBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 56.9 | |
| No-CoTBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 56.8 | |
| SCAFFOLDBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 56.7 | |
| CCoTBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 55.3 | |
| No-CoTBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 55.2 | |
| DDCoTBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 54.8 | |
| MMCoTBackbone=Qwen2.5-VL-32B, Shot=1-shot2025.09 | 54.7 | |
| SCAFFOLDBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 54.2 | |
| MMCoTBackbone=Qwen2.5-VL-32B, Shot=0-shot2025.09 | 53.4 | |
| AIM-CoTBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 46.6 | |
| ICoTBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 46 | |
| DDCoTBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 45.7 | |
| No-CoTBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 45.4 | |
| SCAFFOLDBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 44.9 | |
| AIM-CoTBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 44.7 | |
| CCoTBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 44.1 | |
| ICoTBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 44.1 | |
| No-CoTBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 43.6 | |
| CCoTBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 43.3 | |
| DDCoTBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 42.6 | |
| MMCoTBackbone=Qwen2-VL-7B, Shot=1-shot2025.09 | 42.5 | |
| VRCDdecoding length (L)=384, forward ratio (FR)=0.52026.05 | 41.99 | |
| SCAFFOLDBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 41.7 | |
| AIM-CoTBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 41.5 | |
| MMCoTBackbone=Qwen2-VL-7B, Shot=0-shot2025.09 | 40.1 | |
| VRCDdecoding length (L)=384, forward ratio (FR)=0.252026.05 | 39.82 | |
| AIM-CoTBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 39.7 | |
| ICoTBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 39.4 | |
| IGdecoding length (L)=384, forward ratio (FR)=0.52026.05 | 39.23 | |
| DDCoTBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 38.6 | |
| Margindecoding length (L)=384, forward ratio (FR)=0.52026.05 | 38.47 | |
| Confidencedecoding length (L)=384, forward ratio (FR)=0.52026.05 | 38.35 | |
| CCoTBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 38.2 | |
| Margindecoding length (L)=384, forward ratio (FR)=0.252026.05 | 38.06 | |
| SCAFFOLDBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 38 | |
| No-CoTBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 37.8 | |
| ICoTBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 37.6 | |
| IGdecoding length (L)=384, forward ratio (FR)=0.252026.05 | 37.26 | |
| Confidencedecoding length (L)=384, forward ratio (FR)=0.252026.05 | 37.16 | |
| DDCoTBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 36.8 | |
| MMCoTBackbone=Janus-Pro-7B, Shot=1-shot2025.09 | 36.6 | |
| No-CoTBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 36.5 | |
| CCoTBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 36.4 | |
| SCAFFOLDBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 36.1 | |
| Entropydecoding length (L)=384, forward ratio (FR)=0.52026.05 | 35.86 | |
| MMCoTBackbone=Janus-Pro-7B, Shot=0-shot2025.09 | 35.2 | |
| Entropydecoding length (L)=384, forward ratio (FR)=0.252026.05 | 34.57 | |
| VRCDdecoding length (L)=384, forward ratio (FR)=0.1252026.05 | 34.24 | |
| Margindecoding length (L)=384, forward ratio (FR)=0.1252026.05 | 34.11 | |
| AIM-CoTBackbone=Chameleon-7B, Shot=1-shot2025.09 | 32.8 | |
| ICoTBackbone=Chameleon-7B, Shot=1-shot2025.09 | 32.3 | |
| CCoTBackbone=Chameleon-7B, Shot=1-shot2025.09 | 31.4 | |
| AIM-CoTBackbone=Chameleon-7B, Shot=0-shot2025.09 | 31.4 | |
| SCAFFOLDBackbone=Chameleon-7B, Shot=1-shot2025.09 | 31.1 | |
| MMCoTBackbone=Chameleon-7B, Shot=1-shot2025.09 | 30.6 | |
| DDCoTBackbone=Chameleon-7B, Shot=1-shot2025.09 | 29.8 | |
| ICoTBackbone=Chameleon-7B, Shot=0-shot2025.09 | 29.8 | |
| SCAFFOLDBackbone=Chameleon-7B, Shot=0-shot2025.09 | 29.6 | |
| CCoTBackbone=Chameleon-7B, Shot=0-shot2025.09 | 29.4 | |
| No-CoTBackbone=Chameleon-7B, Shot=0-shot2025.09 | 29.1 | |
| Confidencedecoding length (L)=384, forward ratio (FR)=0.1252026.05 | 28.82 | |
| IGdecoding length (L)=384, forward ratio (FR)=0.1252026.05 | 28.72 | |
| DDCoTBackbone=Chameleon-7B, Shot=0-shot2025.09 | 28.6 | |
| MMCoTBackbone=Chameleon-7B, Shot=0-shot2025.09 | 28.5 | |
| No-CoTBackbone=Chameleon-7B, Shot=1-shot2025.09 | 28.4 | |
| Entropydecoding length (L)=384, forward ratio (FR)=0.1252026.05 | 26.33 |