Reproduce “On the Effectiveness of Weight-Encoded Neural Implicit 3D 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
Thomas Davies, Derek Nowrouzezahrai, Alec Jacobson
The authors encode each 3D shape in a small network whose weights map query points to signed distance, without a separate shape-specific latent code. They train one network per mesh, making the zero-valued surface a continuous, queryable representation. On 10,000 Thingi10k meshes, storage fell from 38.85 GB to 590 MB; this lossy result does not guarantee accuracy for every shape.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.