Video Moment Retrieval on StoryTR (test)
80.4Precision @0.3Gemini-3.0-Pro
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gemini-3.0-ProModel Category=Closed-source Models, Role=Teacher model for data generation2026.04 | 80.4 | 69.8 | 47.4 | 65.9 | 86.6 | 79.5 | 55.1 | 71.8 | 75.7 | 58 | 24.7 | 53.2 | |
| Gemini-2.5-ProModel Category=Closed-source Models2026.04 | 76.6 | 66.5 | 46.2 | 63 | 84 | 78.9 | 54.1 | 70.6 | 71.9 | 56.2 | 23.3 | 50.5 | |
| Shorts-MomentModel Category=Ours, Training=SFT, Backbone=ARC-Hunyuan2026.04 | 63.1 | 52.1 | 30.9 | 49.3 | 69.8 | 62.2 | 42 | 57 | 57.2 | 42 | 15.8 | 39.6 | |
| ARC-HunyuanModel Category=Open-source Baselines, Architecture=Dense 7B2026.04 | 55.1 | 45.7 | 28.1 | 44.7 | 64.7 | 59.9 | 40.6 | 54.2 | 48.7 | 35.3 | 12.7 | 34.4 | |
| Gemini-2.5-FlashModel Category=Closed-source Models2026.04 | 39.4 | 36.5 | 27.6 | 34.4 | 36.9 | 31 | 18.1 | 29.3 | 33.1 | 24.5 | 8.5 | 23.1 | |
| Qwen3-OmniModel Category=Open-source Baselines, Architecture=30B Mixture-of-Experts2026.04 | 15.2 | 12.7 | 8.9 | 13 | 12.7 | 9.8 | 4.7 | 9.9 | 11 | 6.5 | 1.3 | 7.4 |