Fuyang Huang, Ailing Zeng, Minhao Liu, Qiuxia Lai, Qiang Xu
In this paper, we propose a two-stage fully 3D network, namely \textbf{DeepFuse}, to estimate human pose in 3D space by fusing body-worn Inertial Measurement Unit (IMU) data and multi-view images deeply. The first stage is designed for pure vision estimation. To preserve data primitiveness of multi-view inputs, the vision stage uses multi-channel volume as data representation and 3D soft-argmax as activation layer. The second one is the IMU refinement stage which introduces an IMU-bone layer to fuse the IMU and vision data earlier at data level. without requiring a given skeleton model a priori, we can achieve a mean joint error of $28.9$mm on TotalCapture dataset and $13.4$mm on Human3.6M dataset under protocol 1, improving the SOTA result by a large margin. Finally, we discuss the effectiveness of a fully 3D network for 3D pose estimation experimentally which may benefit future research.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| 3D Human Pose Estimation | Total Capture | Average MPJPE (mm) | 28.9 | DeepFuse-IMU |
| 3D Human Pose Estimation | Total Capture | Average MPJPE (mm) | 32.7 | DeepFuse-Vision Only |
| Pose Estimation | Total Capture | Average MPJPE (mm) | 28.9 | DeepFuse-IMU |
| Pose Estimation | Total Capture | Average MPJPE (mm) | 32.7 | DeepFuse-Vision Only |
| 3D | Total Capture | Average MPJPE (mm) | 28.9 | DeepFuse-IMU |
| 3D | Total Capture | Average MPJPE (mm) | 32.7 | DeepFuse-Vision Only |
| 1 Image, 2*2 Stitchi | Total Capture | Average MPJPE (mm) | 28.9 | DeepFuse-IMU |
| 1 Image, 2*2 Stitchi | Total Capture | Average MPJPE (mm) | 32.7 | DeepFuse-Vision Only |