Lightweight Multi-View 3D Pose Estimation through Camera-Disentangled Representation
About
We present a lightweight solution to recover 3D pose from multi-view images captured with spatially calibrated cameras. Building upon recent advances in interpretable representation learning, we exploit 3D geometry to fuse input images into a unified latent representation of pose, which is disentangled from camera view-points. This allows us to reason effectively about 3D pose across different views without using compute-intensive volumetric grids. Our architecture then conditions the learned representation on camera projection operators to produce accurate per-view 2d detections, that can be simply lifted to 3D via a differentiable Direct Linear Transform (DLT) layer. In order to do it efficiently, we propose a novel implementation of DLT that is orders of magnitude faster on GPU architectures than standard SVD-based triangulation methods. We evaluate our approach on two large-scale human pose datasets (H36M and Total Capture): our method outperforms or performs comparably to the state-of-the-art volumetric methods, while, unlike them, yielding real-time performance.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| 3D Human Pose Estimation | Human3.6M (test) | MPJPE (Average)30.2 | 547 | |
| 3D Human Pose Estimation | Human3.6M (subjects 9 and 11) | Average Error30.2 | 180 | |
| 3D Human Pose Estimation | Human3.6M | MPJPE30.2 | 160 | |
| 3D Human Pose Estimation | Human3.6M (S9, S11) | Average Error (MPJPE Avg)30.2 | 94 | |
| 3D Human Pose Estimation | TotalCapture (Seen Cameras (1,3,5,7), Seen Subjects (S1, S2, S3)) | W210.6 | 17 | |
| 3D Human Pose Estimation | TotalCapture (Seen Cameras (1,3,5,7), Unseen Subjects (S4, S5)) | W2 Error27 | 17 | |
| 3D Human Pose Estimation | TotalCapture (Unseen Cameras (2,4,6,8), Seen Subjects (S1, S2)) | W222.4 | 16 | |
| 3D Human Pose Estimation | TotalCapture (Unseen Cameras (2,4,6,8), Overall) | MPJPE38.2 | 16 | |
| 3D Human Pose Estimation | TotalCapture (Unseen Cameras (2,4,6,8), Unseen Subjects (S4, S5)) | Error Metric W239.1 | 16 |