Activity Recognition on H3.6M (Seen)
100AccuracyResNet-50 + Kinetics
Evaluation Results
| Method | Links | |
|---|---|---|
| ResNet-50 + KineticsBackbone=ResNet-50, Pre-training=Kinetics2021.03 | 100 | |
| Ground Truth 3D PoseRepresentation Type=3D Pose Based2021.03 | 100 | |
| NPL + Multi-view projectionsRepresentation Type=Latent 3D Representation, multi-view projections=true2021.03 | 99.7 | |
| NPLRepresentation Type=Latent 3D Representation2021.03 | 99.3 | |
| MultiView 2D PoseRepresentation Type=3D Pose Based2021.03 | 98.3 | |
| Estimated 3D PoseRepresentation Type=3D Pose Based2021.03 | 97.8 | |
| ResNet-50Backbone=ResNet-50, Pre-training=From scratch2021.03 | 86.4 | |
| RandomRepresentation Type=Random baseline2021.03 | 9.1 |