Description-guided Temporal Localization on ShotFinder
85.1ShotHuman
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Human2026.01 | 85.1 | 91.6 | 91.4 | 83.3 | 91.7 | 87.5 | 88.5 | |
| GPT-5.2Model Category=Closed-source, Sampling Strategy=Duration-based (64/128/192 frames), Temperature=02026.01 | 25.5 | 35.5 | 15.7 | 32.5 | 26 | — | 26.9 | |
| Gemini-2.5-ProModel Category=Closed-source, Sampling Strategy=Duration-based (64/128/192 frames), Temperature=02026.01 | 23.5 | 29.5 | 12.9 | 17.5 | 18.5 | 26.5 | 21.3 | |
| Gemini-3-ProModel Category=Closed-source, Sampling Strategy=Duration-based (64/128/192 frames), Temperature=02026.01 | 22.5 | 31 | 15.7 | 26.5 | 21 | 30 | 24.4 | |
| Qwen3-Omni-30B-A3BModel Category=Open-source, Sampling Strategy=Uniform (64 frames), Temperature=02026.01 | 19 | 33 | 11.4 | 18.5 | 17.5 | 21 | 20 | |
| Qwen3-VL-235B-A22BModel Category=Open-source, Sampling Strategy=Uniform (64 frames), Temperature=02026.01 | 18.5 | 28.5 | 20.5 | 17 | 18.5 | — | 20.6 | |
| Claude-4.0-SonnetModel Category=Closed-source, Sampling Strategy=Duration-based (64/128/192 frames), Temperature=02026.01 | 18 | 30 | 8.1 | 15.5 | 19.5 | — | 18.1 | |
| GPT-5-miniModel Category=Closed-source, Sampling Strategy=Duration-based (64/128/192 frames), Temperature=02026.01 | 12 | 21 | 10.5 | 17 | 16 | — | 15.2 |