Revisiting Skeleton-based Action Recognition

Benchmark Model Rank Results
3d-action-recognition-on-assembly101RGBPoseConv3D#3Actions Top-1: 33.61Verbs Top-1: 61.99Object Top-1: 42.90
action-recognition-in-videos-on-ntu-rgbd-120PoseC3D (RGB + Pose)#2Accuracy (Cross-Setup): 96.4Accuracy (Cross-Subject): 95.3
action-recognition-in-videos-on-ntu-rgbd-60PoseC3D (RGB + Pose)#2Accuracy (CS): 97.0Accuracy (CV): 99.6
action-recognition-in-videos-on-volleyballPoseC3D (Pose Only)#1Accuracy: 91.3
action-recognition-on-h2o-2-hands-and-objectsRGBPoseConv3D#7Actions Top-1: 83.47RGB: YesHand Pose: 2DObject Pose: No
group-activity-recognition-on-volleyballPoseC3D (Pose-Only)#5Accuracy: 91.3
skeleton-based-action-recognition-on-kineticsPoseC3D (SlowOnly-346)#2Accuracy: 49.1
skeleton-based-action-recognition-on-kineticsPoseC3D#3Accuracy: 47.7
skeleton-based-action-recognition-on-ntu-rgbd-120PoseC3D (w. HRNet 2D Skeleton)#37Accuracy (Cross-Subject): 86.9Accuracy (Cross-Setup): 90.3
skeleton-based-action-recognition-on-ntu-rgbd-60PoseC3D [3D Heatmap]#5Accuracy (CS): 94.1Accuracy (CV): 97.1Ensembled Modalities: 2