Temporal Grounding on RexTime
38.8R@0.3Qwen3-VL-4B + SynRL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-VL-4B + SynRLParameters=4B2026.03 | 38.8 | 26.2 | 16.3 | 28.9 | |
| VideoChat-TPO2026.03 | 34.5 | 19.3 | 9.8 | 25.2 | |
| EvoGroundTraining Strategy=RL-based2026.05 | 33.5 | 22.4 | — | 25.5 | |
| Time-R1Training Strategy=RL-based, Reproduced=true2026.05 | 32.2 | 22.1 | — | 24.1 | |
| LITA-13BTraining Strategy=SFT-based2026.05 | 29.4 | 16.2 | — | 21.4 | |
| VTimeLLM2026.03 | 28.8 | 17.4 | 7.2 | 20.1 | |
| VTimeLLMTraining Strategy=SFT-based2026.05 | 28.8 | 17.4 | — | 20.1 | |
| VideoChat-R1.5Training Strategy=RL-based, Reproduced=true2026.05 | 28.3 | 18 | — | 20.5 | |
| Qwen3-VL-4BParameters=4B2026.03 | 26.2 | 20.1 | 12.9 | 20.9 | |
| Qwen3-VL-4B CoTParameters=4B, Chain-of-Thought=true2026.03 | 25.4 | 18 | 12.6 | 20.1 | |
| Qwen2.5-VL-32BModel Family=Qwen2.5-VL, Parameters=32B2026.03 | 16.8 | 10 | 5.1 | 13 | |
| Qwen2.5-VLTraining Strategy=SFT-based, Reproduced=true2026.05 | 16.3 | 11.2 | — | 13.3 | |
| TimeChat2026.03 | 14.4 | 7.6 | 3.1 | 11.7 | |
| TimeChatTraining Strategy=SFT-based2026.05 | 14.4 | 7.6 | — | 11.6 | |
| Qwen2.5-VL-7BModel Family=Qwen2.5-VL, Parameters=7B2026.03 | 10.3 | 6.1 | 3 | 8.1 |