Reproduce “NPMs: Neural Parametric Models for 3D Deformable Shapes” 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

Pablo Palafox, Aljaž Božič, Justus Thies, Matthias Nießner, Angela Dai

NPMs: Neural Parametric Models for 3D Deformable Shapes introduces a framework for learning disentangled shape and pose representations of deformable objects directly from 4D data. The approach, from the Technical University of Munich, leverages implicit neural functions to capture fine-scale details and enables accurate reconstruction and tracking of dynamic sequences through test-time optimization.

Key insights

  • High-fidelity, controllable parametric models for deformable shapes can be learned entirely from data, bypassing traditional manual construction and fixed-topology limitations.
  • Implicit neural representations can effectively capture fine-scale details and arbitrary topologies for deformable objects while disentangling shape from pose.
  • Test-time optimization of learned latent codes, combined with encoder-based initialization, enables robust and accurate fitting of implicit deformable models to sparse monocular observations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.