Weakly-Supervised 3D Human Pose Learning via Multi-view Images in the Wild

CVPR 2020  ·  Umar Iqbal, Pavlo Molchanov, Jan Kautz ·

One major challenge for monocular 3D human pose estimation in-the-wild is the acquisition of training data that contains unconstrained images annotated with accurate 3D poses. In this paper, we address this challenge by proposing a weakly-supervised approach that does not require 3D annotations and learns to estimate 3D poses from unlabeled multi-view data, which can be acquired easily in in-the-wild environments. We propose a novel end-to-end learning framework that enables weakly-supervised training using multi-view consistency. Since multi-view consistency is prone to degenerated solutions, we adopt a 2.5D pose representation and propose a novel objective function that can only be minimized when the predictions of the trained model are consistent and plausible across all camera views. We evaluate our proposed approach on two large scale datasets (Human3.6M and MPII-INF-3DHP) where it achieves state-of-the-art performance among semi-/weakly-supervised methods.

PDF Abstract CVPR 2020 PDF CVPR 2020 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Weakly-supervised 3D Human Pose Estimation Human3.6M GeoRep Average MPJPE (mm) 67.4 # 18
Weakly-supervised 3D Human Pose Estimation Human3.6M GeoRep (semi-supervised) Average MPJPE (mm) 59.7 # 10
3D Human Pose Estimation Human3.6M GeoRep (fully-supervised) Average MPJPE (mm) 56.1 # 222
Weakly-supervised 3D Human Pose Estimation MPI-INF-3DHP GeoRep (semi-supervised) MPJPE 113.8 # 1
PCK 79.1 # 1
Weakly-supervised 3D Human Pose Estimation MPI-INF-3DHP GeoRep MPJPE 122.4 # 3
PCK 76.5 # 2
3D Human Pose Estimation MPI-INF-3DHP GeoRep (fully-supervised) MPJPE 110.8 # 77
PCK 80.2 # 61

Methods


No methods listed for this paper. Add relevant methods here