Dynamic Scene VQA on RailVQA-bench
82.53MCQ AccuracyRailVQA-CoM
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| RailVQA-CoMLMM=Qwen3-VL2026.03 | 82.53 | 74 | 80.6 | 68.4 | 84.8 | 66.5 | 14.88 | 13.88 | 13.47 | |
| LMM OnlyLMM=Qwen3-VL2026.03 | 82.36 | 71.01 | 78.9 | 63.8 | 79.6 | 54.8 | 23.21 | 8.16 | 7.96 | |
| Unadapted DetectorLMM=Qwen3-VL2026.03 | 82.17 | 71.8 | 81.3 | 64.7 | 85.4 | 58.9 | 15.01 | 13.18 | 13.06 | |
| RailVQA-CoMLMM=InternVL3.52026.03 | 74.8 | 66.73 | 75.3 | 58.8 | 80.1 | 59.6 | 21.86 | 11.54 | 10.4 | |
| LMM OnlyLMM=InternVL3.52026.03 | 67.16 | 60.6 | 69.4 | 54.4 | 71.6 | 53 | 28.71 | 5.45 | 5.8 | |
| Unadapted DetectorLMM=InternVL3.52026.03 | 60.45 | 52.5 | 67.6 | 41.1 | 71.5 | 43.9 | 22.01 | 11.32 | 10.28 |