Color Understanding on ColorBench
42.2AccuracyJigsaw + CARE
Evaluation Results
| Method | Links | |
|---|---|---|
| Jigsaw + CARE2025.12 | 42.2 | |
| Jigsaw + CL2025.12 | 41.26 | |
| Mix + CL + CARE2025.12 | 40.84 | |
| Rotation + CL + CARE2025.12 | 40.73 | |
| Visual Jigsaw2025.12 | 40.21 | |
| VisualSphinx2025.12 | 39.9 | |
| Jigsaw + CL + CARE2025.12 | 39.69 | |
| ViCrit2025.12 | 39.06 | |
| Qwen 2.5 VL2025.12 | 37.38 | |
| Jigsaw2025.12 | 37.07 | |
| JigsawPrompting=CoT prompting, Backbone=Qwen-VL-2.5 3B2025.12 | 36.96 | |
| VLM-R1Prompting=CoT prompting, Backbone=Qwen-VL-2.5 3B2025.12 | 36.02 | |
| MixPrompting=CoT prompting, Backbone=Qwen-VL-2.5 3B2025.12 | 36.02 | |
| Vision-Zero2025.12 | 35.81 | |
| GRPO-CARE2025.12 | 35.39 | |
| PatchFit + CL + CARE2025.12 | 34.66 | |
| Jigsaw-R1Prompting=CoT prompting, Backbone=Qwen-VL-2.5 3B2025.12 | 32.25 | |
| Qwen 2.5 VLPrompting=CoT prompting, Backbone=Qwen-VL-2.5 3B2025.12 | 31.51 |