Learning to Reconstruct 3D Human Pose and Shape via Model-fitting in the Loop

Model-based human pose estimation is currently approached through two different paradigms. Optimization-based methods fit a parametric body model to 2D observations in an iterative manner, leading to accurate image-model alignments, but are often slow and sensitive to the initialization. In contrast, regression-based methods, that use a deep network to directly estimate the model parameters from pixels, tend to provide reasonable, but not pixel accurate, results while requiring huge amounts of supervision. In this work, instead of investigating which approach is better, our key insight is that the two paradigms can form a strong collaboration. A reasonable, directly regressed estimate from the network can initialize the iterative optimization making the fitting faster and more accurate. Similarly, a pixel accurate fit from iterative optimization can act as strong supervision for the network. This is the core of our proposed approach SPIN (SMPL oPtimization IN the loop). The deep network initializes an iterative optimization routine that fits the body model to 2D joints within the training loop, and the fitted estimate is subsequently used to supervise the network. Our approach is self-improving by nature, since better network estimates can lead the optimization to better solutions, while more accurate optimization fits provide better supervision for the network. We demonstrate the effectiveness of our approach in different settings, where 3D ground truth is scarce, or not available, and we consistently outperform the state-of-the-art model-based pose estimation approaches by significant margins. The project website with videos, results, and code can be found at https://seas.upenn.edu/~nkolot/projects/spin.

PDF Abstract ICCV 2019 PDF ICCV 2019 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
3D Human Pose Estimation 3D Poses in the Wild Challenge SPIN MPJPE 102.56 # 5
MPJAE 25.42 # 3
3D Human Pose Estimation 3DPW SPIN PA-MPJPE 59.2 # 94
MPJPE 96.9 # 100
3D Multi-Person Pose Estimation AGORA SPIN B-NMVE 216.3 # 3
B-NMJE 223.1 # 3
B-MVE 168.7 # 3
B-MPJPE 175.1 # 3
3D Human Pose Estimation AGORA SPIN B-NMVE 216.3 # 10
B-NMJE 223.1 # 10
B-MVE 168.7 # 10
B-MPJPE 175.1 # 10
3D Human Pose Estimation Human3.6M SPIN PA-MPJPE 41.1 # 60
3D Human Pose Estimation MPI-INF-3DHP SPIN AUC 37.1 # 71
MPJPE 105.2 # 74
PCK 76.4 # 72
3D Human Pose Estimation MPI-INF-3DHP SPIN (Rigid Alignment) AUC 55.6 # 31
PA-MPJPE 67.5 # 22
PCK 92.5 # 23

Results from Other Papers

Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
3D Human Shape Estimation SSP-3D SPIN PVE-T-SC 22.2 # 8
mIOU 70.0 # 4


No methods listed for this paper. Add relevant methods here