Release v0.4.1

Features

• Added tmesh2voxel and tmesh2sparse conversion functions for constructing voxel grids directly from TriangleMesh and computing active voxels based on exact SDF.
• Implemented extremely fast, highly parallel CUDA Frontier-based BFS kernels to perform fix_winding and fix_inversion directly on the GPU.
• Added TriangleMesh::fix_normals() logic in C++ to execute the GPU kernel passes, repairing non-manifold and misoriented polygon-soup normal topologies to create watertight implicit surfaces.
• Exposed fix_normals from the native TriangleMesh via Pybind11.

Refactors & Build

• Upgraded the package version to 0.4.1.
• Added missing Mesa and OpenGL dependencies in Dockerfile and isolated heavy pip installation dependencies.
• Added build fallback logic to populate nvcc_flags when a GPU is not present during package compilation.
• Refactored main.py in the marching cubes pipeline to entirely remove Kaolin dependencies in favor of native TriangleMesh bindings.

Examples

• Added main.sh evaluation script in experiments/marching_cubes to automate Marching Cubes extractions across all 25 included 3D assets and 5 target spatial resolutions.