Unsupervised 3D Pose Estimation with Geometric Self-Supervision
This addresses the problem of expensive 3D data annotation for pose estimation researchers, representing a novel unsupervised approach rather than incremental.
The paper tackles unsupervised 3D human pose estimation from single 2D images without requiring multi-view data or 3D annotations, achieving a 30% improvement over previous unsupervised methods on the Human3.6M dataset and outperforming many weakly supervised approaches.
We present an unsupervised learning approach to recover 3D human pose from 2D skeletal joints extracted from a single image. Our method does not require any multi-view image data, 3D skeletons, correspondences between 2D-3D points, or use previously learned 3D priors during training. A lifting network accepts 2D landmarks as inputs and generates a corresponding 3D skeleton estimate. During training, the recovered 3D skeleton is reprojected on random camera viewpoints to generate new "synthetic" 2D poses. By lifting the synthetic 2D poses back to 3D and re-projecting them in the original camera view, we can define self-consistency loss both in 3D and in 2D. The training can thus be self supervised by exploiting the geometric self-consistency of the lift-reproject-lift process. We show that self-consistency alone is not sufficient to generate realistic skeletons, however adding a 2D pose discriminator enables the lifter to output valid 3D poses. Additionally, to learn from 2D poses "in the wild", we train an unsupervised 2D domain adapter network to allow for an expansion of 2D data. This improves results and demonstrates the usefulness of 2D pose data for unsupervised 3D lifting. Results on Human3.6M dataset for 3D human pose estimation demonstrate that our approach improves upon the previous unsupervised methods by 30% and outperforms many weakly supervised approaches that explicitly use 3D data.