Polar Coordinate-Based 2D Pose Prior with Neural Distance Field
Qi Gan, Sao Mai Nguyen, Eric Fenaux, Stephan Clémençon, Mounîm El Yacoubi
TL;DR
This work introduces a polar-coordinate 2D pose representation for a Neural Distance Field (NDF) prior to refine 2D human poses in sports settings. By explicitly encoding joint lengths and using an arc-radius distance, the method corrects both angular and radial pose errors with a gradient-based data augmentation strategy that enables training from limited data. The approach improves pose plausibility across multiple pose representations on a long-jump dataset, demonstrating robustness to domain shifts while highlighting convergence and large-motion joint challenges as directions for future work. Overall, the polar NDF prior provides a data-efficient, domain-robust refinement mechanism for 2D HPE in sports contexts, with code available online.
Abstract
Human pose capture is essential for sports analysis, enabling precise evaluation of athletes' movements. While deep learning-based human pose estimation (HPE) models from RGB videos have achieved impressive performance on public datasets, their effectiveness in real-world sports scenarios is often hindered by motion blur, occlusions, and domain shifts across different pose representations. Fine-tuning these models can partially alleviate such challenges but typically requires large-scale annotated data and still struggles to generalize across diverse sports environments. To address these limitations, we propose a 2D pose prior-guided refinement approach based on Neural Distance Fields (NDF). Unlike existing approaches that rely solely on angular representations of human poses, we introduce a polar coordinate-based representation that explicitly incorporates joint connection lengths, enabling a more accurate correction of erroneous pose estimations. Additionally, we define a novel non-geodesic distance metric that separates angular and radial discrepancies, which we demonstrate is better suited for polar representations than traditional geodesic distances. To mitigate data scarcity, we develop a gradient-based batch-projection augmentation strategy, which synthesizes realistic pose samples through iterative refinement. Our method is evaluated on a long jump dataset, demonstrating its ability to improve 2D pose estimation across multiple pose representations, making it robust across different domains. Experimental results show that our approach enhances pose plausibility while requiring only limited training data. Code is available at: https://github.com/QGAN2019/polar-NDF.