Reproduce “NURBS-Diff: A Differentiable Programming Module for NURBS” 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
Anjana Deva Prasad, Aditya Balu, Harshil Shah, Soumik Sarkar, Chinmay Hegde, Adarsh Krishnamurthy
NURBS-Diff is a differentiable programming module that integrates Non-Uniform Rational B-splines (NURBS) directly into deep learning frameworks. It enables the optimization of all NURBS parameters, including knot vectors via a Gaussian smoothing approximation for derivatives, demonstrating superior unsupervised point cloud reconstruction with an order of magnitude improvement in Chamfer Distance and robust enforcement of geometric constraints in engineering applications.
Key insights
- Comprehensive differentiability of all NURBS parameters, including control points, weights, and knot vectors, allows for direct and flexible geometric optimization within deep learning.
- Approximating derivatives of basis functions with respect to knot vectors via Gaussian smoothing provides a practical method for gradient-based reparameterization of NURBS surfaces.
- Direct integration of NURBS into deep learning preserves geometric fidelity and design intent, enabling end-to-end CAD-aware deep learning applications without data conversion compromises.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.