Temporal language grounding on QVHighlights (test)
40.67mAP (Avg)DemaFormer
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DemaFormerPre-training=ASR captions2023.12 | 40.67 | 63.55 | 45.87 | 59.32 | 41.94 | 65.03 | |
| DemaFormerPre-training=false2023.12 | 38.71 | 62.39 | 43.94 | 58.25 | 39.36 | 64.77 | |
| UMTPre-training=ASR captions2023.12 | 38.08 | 60.83 | 43.26 | 57.33 | 39.12 | 62.39 | |
| READ#params (M)=0.19 (1.20%)2023.12 | 36.74 | — | — | — | — | — | |
| Moment-DETRPre-training=ASR captions2023.12 | 36.14 | 59.78 | 40.33 | 60.51 | 35.36 | 60.17 | |
| Full#params (M)=15.88 (100%)2023.12 | 36.14 | — | — | — | — | — | |
| UMTPre-training=false2023.12 | 36.12 | 56.23 | 41.18 | 53.83 | 37.01 | 59.99 | |
| XML+Pre-training=false2023.12 | 34.9 | 46.69 | 33.46 | 47.89 | 34.67 | 55.06 | |
| LoRA#params (M)=0.19 (1.20%)2023.12 | 33.96 | — | — | — | — | — | |
| Adapter#params (M)=0.20 (1.26%)2023.12 | 33.61 | — | — | — | — | — | |
| Proj#params (M)=7.31 (46.03%)2023.12 | 32.71 | — | — | — | — | — | |
| XMLPre-training=false2023.12 | 32.14 | 41.83 | 30.35 | 44.63 | 31.73 | 55.25 | |
| Moment-DETRPre-training=false2023.12 | 30.73 | 52.89 | 33.02 | 54.82 | 29.4 | 55.6 | |
| Partial#params (M)=1.05 (6.61%)2023.12 | 26.37 | — | — | — | — | — | |
| Prompt#params (M)=0.04 (0.25%)2023.12 | 25.86 | — | — | — | — | — | |
| Bias#params (M)=0.06 (0.38%)2023.12 | 24.89 | — | — | — | — | — | |
| CLIPPre-training=false2023.12 | 7.67 | 16.88 | 5.19 | 18.11 | 7 | 61.04 |