Version 1.0

Legend for changelogs

• Feature something that you couldn’t do before.

• Efficiency an existing feature now may not require as much computation or memory.

• Enhancement a miscellaneous minor improvement.

• Fix something that previously didn’t work as documented – or according to reasonable expectations – should now work.

• API Change you will need to change your code to have the same effect in the future; or a feature will be removed in the future.

Version 1.0.1

May 2026

Released new version because of naming conflict in pypi.

Version 1.0.0

May 2026

First version of DiGeo, with core functionalities for geodesic computations, mesh handling, and differentiable optimization on meshes. By Hippolyte Verninas, Caner Korkmaz, Stefanos Zafeiriou, Tolga Birdal and Simone Foti.

digeo

• Feature digeo.Mesh and digeo.MeshBatch to represent triangle meshes and batches of meshes on CPU and GPU.

• Feature digeo.load_mesh_from_file() and digeo.load_mesh_from_trimesh() to load meshes from disk or from Trimesh objects.

• Feature digeo.MeshPoint and digeo.MeshPointBatch to represent points on meshes as face indices and barycentric coordinates.

digeo.ops

• Feature ops.trace_geodesics() to trace straightest geodesics (exponential map) on meshes, with multiple gradient computation strategies: "none", "ep" (Extrinsic Proxy), "gfd" (Geodesic Finite Differences) and "abfd" (Adjoint Backward Finite Differences). See Differentiation methods for details.

• Feature ops.uniform_sampling() to uniformly sample points on the surface of a mesh. Fixed by Kim-Chan Tze Hui, Louiz.

digeo.nn

• Feature nn.BiharmonicDistance to compute biharmonic distances between points on a mesh.

• Feature nn.AGC (Adaptive Geodesic Convolution) layer that performs convolution over geodesic polar patches on the mesh surface, suitable for geometric deep learning tasks.

digeo.optim

• Feature optim.mesh_gd() to perform Riemannian gradient descent on meshes, with optional Wolfe line search.

• Feature optim.mesh_lbfgs() to execute the Mesh-LBFGS Riemannian optimization algorithm, which adapts L-BFGS to the mesh manifold via parallel transport. See the Voronoi Tessellation example for an application to geodesic centroidal Voronoi tessellation.