Multi-region two-stream R-CNN for action detection

Benchmark Model Rank Results
action-detection-on-j-hmdbMR-TS R-CNNFrame-mAP 0.5: 58.5Video-mAP 0.2: 74.3Video-mAP 0.5: 73.09
action-detection-on-j-hmdbTS R-CNNFrame-mAP 0.5: 56.9Video-mAP 0.2: 71.1Video-mAP 0.5: 70.6
action-detection-on-ucf101-24MR-TS R-CNNFrame-mAP 0.5: 39.63
action-detection-on-ucf101-24TS R-CNNFrame-mAP 0.5: 39.94
action-recognition-in-videos-on-ucf101MR Two-Sream R-CNN3-fold Accuracy: 91.1
skeleton-based-action-recognition-on-j-hmdbMR Two-Sream R-CNNAccuracy (RGB+pose): 71.1