Visual Understanding on RealWorldQA
68.23Accuracy (Clean)Robust-R1 (SFT)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Robust-R1 (SFT)Category=Ours, Training Mode=SFT2025.12 | 68.23 | 67.58 | 67.32 | 63.92 | |
| Robust-R1 (SFT and RL)Category=Ours, Training Mode=SFT and RL2025.12 | 67.71 | 66.4 | 67.05 | 63.26 | |
| Qwen2.5-VL-3BCategory=General MLLM2025.12 | 65.22 | 64.96 | 63.39 | 60.65 | |
| InternVL-4BCategory=General MLLM2025.12 | 57.38 | 58.16 | 57.64 | 54.9 | |
| Gemma3-4BCategory=General MLLM2025.12 | 55.42 | 54.77 | 53.72 | 52.81 | |
| Robust CLIPCategory=Robust MLLM2025.12 | 43.26 | 42.48 | 42.61 | 41.43 | |
| TeCoACategory=Robust MLLM2025.12 | 40 | 39.73 | 39.47 | 38.69 |