Hierarchical Self-Attention Network for Action Localization in Videos

Benchmark Model Rank Results
action-detection-on-j-hmdbHISAN (ResNet-101 + FPN)Video-mAP 0.2: 87.59Video-mAP 0.5: 86.49
action-detection-on-j-hmdbHISAN (VGG-16)Frame-mAP 0.5: 76.72Video-mAP 0.2: 85.97Video-mAP 0.5: 84.02
action-detection-on-ucf101-24HISAN (ResNet-101 + FPN)Video-mAP 0.2: 82.30Video-mAP 0.5: 51.47
action-detection-on-ucf101-24HISAN (VGG-16)Frame-mAP 0.5: 73.71Video-mAP 0.2: 80.42Video-mAP 0.5: 49.50