SEAL-pose: Enhancing 3D Human Pose Estimation via a Learned Loss for Structural Consistency
This work improves 3D human pose estimation for applications like motion capture and robotics by offering a more effective, data-driven alternative to manual structural constraints, though it is incremental as it builds on existing backbone methods.
The paper tackles the problem of 3D human pose estimation by addressing limitations in capturing joint dependencies with conventional supervised losses, proposing SEAL-pose, a data-driven framework that uses a learnable loss-net to train a pose-net for structural consistency, resulting in reduced per-joint errors and improved pose plausibility across multiple benchmarks and backbones.
3D human pose estimation (HPE) is characterized by intricate local and global dependencies among joints. Conventional supervised losses are limited in capturing these correlations because they treat each joint independently. Previous studies have attempted to promote structural consistency through manually designed priors or rule-based constraints; however, these approaches typically require manual specification and are often non-differentiable, limiting their use as end-to-end training objectives. We propose SEAL-pose, a data-driven framework in which a learnable loss-net trains a pose-net by evaluating structural plausibility. Rather than relying on hand-crafted priors, our joint-graph-based design enables the loss-net to learn complex structural dependencies directly from data. Extensive experiments on three 3D HPE benchmarks with eight backbones show that SEAL-pose reduces per-joint errors and improves pose plausibility compared with the corresponding backbones across all settings. Beyond improving each backbone, SEAL-pose also outperforms models with explicit structural constraints, despite not enforcing any such constraints. Finally, we analyze the relationship between the loss-net and structural consistency, and evaluate SEAL-pose in cross-dataset and in-the-wild settings.