3D Hand Pose Estimation on ICVL (test)
4.76Mean Error (mm)Virtual View Selection and Fusion Module
Evaluation Results
| Method | Links | |
|---|---|---|
| Virtual View Selection and Fusion Modulenumber of views=152022.03 | 4.76 | |
| Virtual View Selection and Fusion Modulenumber of views=92022.03 | 4.77 | |
| Virtual View Selection and Fusion Modulenumber of views=252022.03 | 4.79 | |
| Virtual View Selection and Fusion Modulenumber of views=32022.03 | 4.86 | |
| Virtual View Selection and Fusion Modulenumber of views=12022.03 | 5.16 | |
| HandFoldingNetPercentage of ground-truth annotations used for training=100%2023.03 | 5.95 | |
| HandFoldingNetInput=3D, Type=Regression-based (R)2021.08 | 5.95 | |
| AWR2022.03 | 5.98 | |
| OursPercentage of ground-truth annotations used for training=100%2023.03 | 5.99 | |
| OursLabel Usage=100%, Augmented Set=No2023.03 | 5.99 | |
| JGR-P2OPercentage of ground-truth annotations used for training=100%2023.03 | 6.02 | |
| JGR-P2OInput=2D, Type=Detection-based (D)2021.08 | 6.02 | |
| OursLabel Usage=75%, Augmented Set=No2023.03 | 6.04 | |
| OursLabel Usage=50%, Augmented Set=No2023.03 | 6.06 | |
| OursPercentage of ground-truth annotations used for training=25%2023.03 | 6.11 | |
| OursLabel Usage=25%, Augmented Set=No2023.03 | 6.11 | |
| V2V-PoseNet2017.11 | 6.28 | |
| V2V2022.03 | 6.28 | |
| V2V-PoseNetPercentage of ground-truth annotations used for training=100%2023.03 | 6.28 | |
| V2VInput=3D, Type=Detection-based (D)2021.08 | 6.28 | |
| V2V*ensemble=10 models, FPS=3.52019.08 | 6.286 | |
| Point-to-PointPercentage of ground-truth annotations used for training=100%2023.03 | 6.3 | |
| Point-to-PointInput=3D, Type=Detection-based (D)2021.08 | 6.3 | |
| P2PFPS=41.82019.08 | 6.328 | |
| P2P2022.03 | 6.33 | |
| A2J2022.03 | 6.46 | |
| A2JPercentage of ground-truth annotations used for training=100%2023.03 | 6.46 | |
| A2JFPS=105.062019.08 | 6.461 | |
| NARHTPercentage of ground-truth annotations used for training=100%2023.03 | 6.47 | |
| CrossInfoNetPercentage of ground-truth annotations used for training=100%2023.03 | 6.73 | |
| CrossInfoNetInput=2D, Type=Regression-based (R)2021.08 | 6.73 | |
| Pose-REN2017.11 | 6.79 | |
| Pose-REN2019.08 | 6.79 | |
| Pose-RenInput=2D, Type=Regression-based (R)2021.08 | 6.79 | |
| HandPointNet2022.03 | 6.93 | |
| HandPointNetFPS=482019.08 | 6.935 | |
| HandPointNetPercentage of ground-truth annotations used for training=100%2023.03 | 6.94 | |
| OursLabel Usage=1%, Augmented Set=No2023.03 | 6.94 | |
| HandPointNetInput=3D, Type=Regression-based (R)2021.08 | 6.94 | |
| LSPSLabel Usage=100%, Augmented Set=No2023.03 | 7 | |
| LSPSLabel Usage=75%, Augmented Set=No2023.03 | 7.05 | |
| LSPSLabel Usage=50%, Augmented Set=No2023.03 | 7.1 | |
| SHPR-NetPercentage of ground-truth annotations used for training=100%2023.03 | 7.22 | |
| SHPR-NetInput=3D, Type=Regression-based (R)2021.08 | 7.22 | |
| DenseReg2022.03 | 7.24 | |
| DenseRegFPS=27.82019.08 | 7.3 | |
| DenseRegPercentage of ground-truth annotations used for training=100%2023.03 | 7.3 | |
| DenseRegInput=2D, Type=Detection-based (D)2021.08 | 7.3 | |
| REN-9x6x62017.11 | 7.31 | |
| REN-9x6x62019.08 | 7.31 | |
| REN-9x6x6Percentage of ground-truth annotations used for training=100%2023.03 | 7.31 | |
| Ren-9x6x6Input=2D, Type=Regression-based (R)2021.08 | 7.31 | |
| LSPSLabel Usage=25%, Augmented Set=No2023.03 | 7.35 | |
| REN-4x6x62017.11 | 7.63 | |
| REN-4x6x62019.08 | 7.63 | |
| REN-4x6x6Percentage of ground-truth annotations used for training=100%2023.03 | 7.63 | |
| Ren-4x6x6Input=2D, Type=Regression-based (R)2021.08 | 7.63 | |
| SO-HandNetLabel Usage=100%, Augmented Set=No2023.03 | 7.7 | |
| DeepPrior++2017.11 | 8.1 | |
| DeepPrior++FPS=302019.08 | 8.1 | |
| DeepPrior++Percentage of ground-truth annotations used for training=100%2023.03 | 8.1 | |
| DeepPrior++Input=2D, Type=Regression-based (R)2021.08 | 8.1 | |
| Baek et al.Label Usage=100%, Augmented Set=Yes, 10 times2023.03 | 8.5 | |
| Baek et al.(w/o refine)Label Usage=100%, Augmented Set=Yes, 10 times2023.03 | 9.1 | |
| SO-HandNetLabel Usage=75%, Augmented Set=No2023.03 | 9.1 | |
| JTSC2017.11 | 9.16 | |
| JTSC2019.08 | 9.16 | |
| SO-HandNetLabel Usage=50%, Augmented Set=No2023.03 | 9.4 | |
| Cascade2017.11 | 9.9 | |
| Cascade2019.08 | 9.9 | |
| Crossing NetLabel Usage=50%, Augmented Set=No2023.03 | 10 | |
| Crossing NetLabel Usage=75%, Augmented Set=No2023.03 | 10.1 | |
| CrossingNets2017.11 | 10.2 | |
| CrossingNetsFPS=90.92019.08 | 10.2 | |
| Crossing NetLabel Usage=100%, Augmented Set=No2023.03 | 10.2 | |
| DeepPrior2017.11 | 10.4 | |
| DeepPriorPercentage of ground-truth annotations used for training=100%2023.03 | 10.4 | |
| Baek et al.(w/o aug.; refine)Label Usage=100%, Augmented Set=No2023.03 | 10.4 | |
| DeepPriorInput=2D, Type=Regression-based (R)2021.08 | 10.4 | |
| CDO2017.11 | 10.5 | |
| Crossing NetLabel Usage=25%, Augmented Set=No2023.03 | 10.5 | |
| Hand3D2017.11 | 10.9 | |
| Hand3DFPS=302019.08 | 10.9 | |
| SO-HandNetLabel Usage=25%, Augmented Set=No2023.03 | 11.1 | |
| DeepModel2017.11 | 11.56 | |
| DeepModel2019.08 | 11.56 | |
| DeepModelPercentage of ground-truth annotations used for training=100%2023.03 | 11.56 | |
| DeepModelInput=2D, Type=Regression-based (R)2021.08 | 11.56 | |
| Baek et al.(baseline)Label Usage=100%, Augmented Set=No2023.03 | 12.1 | |
| LRF2017.11 | 12.58 | |
| LRF2019.08 | 12.58 |