Activity Recognition on MPII (test)
37.5mAPOurs, Res101
Evaluation results
| Method | Links | |
|---|---|---|
| Ours, Res101Full Img=true, Bbox=true, Pose=true2018.07 | 37.5 | |
| Ours, VGG16Full Img=true, Bbox=true, Pose=true2018.07 | 36.8 | |
| Pose Reg. Attn. Pooling, Res101Full Img=true, Bbox=false, Pose=true2018.07 | 36.1 | |
| Pose Regularized Attentional Pooling (R-101)Full Img=true, Pose=true2017.11 | 36.1 | |
| Attentional Pooling (R-101)Full Img=true2017.11 | 36 | |
| Fusion-2Supervision strategy=label per person instance, Bounding box source=ground truth2016.04 | 32.24 | |
| Mallya & Lazebnik, VGG16Full Img=true, Bbox=true, Pose=false2018.07 | 32.2 | |
| VGG16, Fusion (best)Full Img=true, Bbox=true2017.11 | 32.2 | |
| Fusion-1Supervision strategy=label per person instance, Bounding box source=ground truth2016.04 | 32.06 | |
| Fusion-2Supervision strategy=label per person instance, Bounding box source=detected2016.04 | 32.02 | |
| VGG16, Fusion+MIL (best)Full Img=true, Bbox=true, MIL=true2017.11 | 31.9 | |
| Fusion-2Supervision strategy=MIL, Bounding box source=ground truth2016.04 | 31.89 | |
| Fusion-2Supervision strategy=MIL, Bounding box source=detected2016.04 | 31.81 | |
| Fusion-1Supervision strategy=MIL, Bounding box source=ground truth2016.04 | 31.68 | |
| VGG-16, R*CNNBackbone=VGG-162016.04 | 26.7 | |
| R*CNN, VGG16Full Img=false, Bbox=true, Pose=false2018.07 | 26.7 | |
| VGG16, R*CNNFull Img=true, Bbox=true2017.11 | 26.7 | |
| Dense Trajectory + Pose2016.04 | 5.5 | |
| Dense Trajectory + PoseFull Img=true, Bbox=false, Pose=true2018.07 | 5.5 | |
| Dense Trajectory + PosePose=true2017.11 | 5.5 |