Method | Mode | T(IOU@0.3) | T(IOU@0.5) | A(IOU@0.5) |
---|
G-TAD [48] | Full | – | 40.2 | 46.7 |
P-GCN [49] | Full | 63.6 | 49.1 | 48.3 |
Nguyen [36] | Weak | 46.6 | 26.8 | – |
3C-Net [34] | Weak | 40.9 | 24.6 | 35.4 |
WSGN [31] | Weak | 42.0 | 25.1 | – |
Islam [29] | Weak | 46.8 | 29.6 | 35.2 |
BaS-Net [35] | Weak | 44.6 | 27.0 | 34.5 |
DGAM [32] | Weak | 46.8 | 28.8 | 41.0 |
HAM-Net [39] | Weak | 50.3 | 31.0 | 41.5 |
Ours | Weak | 64.4 | 49.6 | 52.2 |
- The table lists the comparison results of mAP (16 frames clip). We compared with typical fully and weakly supervised methods. T(IOU@0.3) indicates THUMOS14 with IOU@0.3, T(IOU@0.5) indicates IOU=0.5, and A(IOU@0.5) indicates ActivityNet with IOU@0.5. Note that, the proposed method is an object location-unsupervised classification-supervised attention network