← All resources

Parallelised Differentiable Straightest Geodesics for 3D Meshes

Auto-discovered

Paper Advanced

Hippolyte Verninas, Caner Korkmaz, Stefanos Zafeiriou, Tolga Birdal, Simone Foti

Visit resource ↗

Description

Parallelised Differentiable Straightest Geodesics for 3D Meshes presents a CIRCLE Group Imperial College London CVPR 2026 paper by Hippolyte Verninas, Caner Korkmaz, Stefanos Zafeiriou, Tolga Birdal, and Simone Foti on differentiable geodesic operators for mesh learning.

The work parallelizes straightest geodesics on GPU and derives Extrinsic Proxy and Geodesic Finite Differences Jacobians so exponential maps, tracing, and parallel transport can backpropagate. Shown uses include Adaptive Geodesic Convolution, MeshFlow on meshes, and Mesh-LBFGS for centroidal Voronoi tessellation.

It is for researchers and advanced practitioners in geometric deep learning on meshes who need differentiable Riemannian operators.

The project page summarizes the method, demos, BibTeX, and the digeo pip package; the paper is the primary technical source.

It is useful as a compact bridge from discrete geodesic theory to trainable mesh pipelines. Link last verified accessible on 2026-09-27. Link last verified accessible on 2026-09-28.