Reproduce “PoseForge: Editable Pose Analytics for AI-Assisted Sports Coaching” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuvam Swapnil Dash, Arpit Narechania
PoseForge combines monocular 3D pose reconstruction, phase-aware cricket biomechanics, interactive inverse-kinematics editing, and language-model feedback in a visual analytics system for batting coaching. Evaluated with 11 cricket experts, it supported frame-level correction analysis and plausible what-if edits, while AI tests achieved 47/50 correct joint resolutions, 45/50 axis selections, 44/50 frame targets, and 23/30 correct coaching responses.
Key insights
- Visualizing both the direction and magnitude of a correction helped experts connect numerical analysis with practical coaching language, while constrained edits increased trust in the simulated postures.
- Experts agreed more consistently on the direction of desirable changes than on their exact magnitude, supporting adjustable, player- or coach-specific norms rather than a single universal batting template.
- The framework connects diagnosis to intervention, but its usefulness depends on contextual factors such as delivery type, pitch conditions, shot situation, viewpoint, and the limitations of monocular pose estimation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.