Online spatio-temporal understanding on OST-Bench
80.8Directional Temporal ScoreViSRA
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ViSRABackbone=InternVL3-8B2026.05 | 80.8 | 63.1 | 71.4 | 78.5 | 69.1 | 73.3 | |
| ViSRABackbone=LLaVA-OneVision-7B2026.05 | 79.9 | 61.4 | 72.2 | 78.1 | 69.4 | 73 | |
| ViSRABackbone=Qwen2.5-VL-7B2026.05 | 66.6 | 62.3 | 63.1 | 70.6 | 66.3 | 67.1 | |
| Qwen2.5-VL-7BType=Base Model2026.05 | 28.3 | 44.4 | 12.8 | 40.5 | 45 | 38.5 | |
| InternVL3-8BType=Base Model2026.05 | 25.5 | 38.5 | 13.1 | 38.2 | 46.5 | 37.2 | |
| LLaVA-OneVision-7BType=Base Model2026.05 | 19.2 | 24.4 | 23.5 | 38.1 | 12.8 | 25.3 |