Release v0.3.9

Features

• Introduced complete Point Transformer architecture (Layer, Block, Cls) optimized for fixed-size continuous coordinate analysis.
• Implemented fast CUDA Z-curve Morton Encoding with full PyTorch z_curve_sort integration.
• Added exact one-sided and bidirectional Hausdorff distance computations.
• Developed PointDigit3D dataloader utilizing CPU-safe trimesh surface point and normal sampling.
• Enhanced marching_tetrahedra optimization with fairness losses, best-checkpoint restorations, and asset batch processing.

Refactors & Build

• Fully migrated all legacy sparse voxel classification and VAE components to the new experiments module structure.
• Hardened C++ compilation by fixing inline linkages and replacing ternary operations in CUDA headers.
• Completely dropped meshio in favor of trimesh for fast OBJ/OFF parsing.
• Added official Dockerfile and docker-compose.yml configurations for streamlined deployments.
• Updated pyproject.toml dependencies and setup scripts.

Examples

• Added detailed Jupyter Notebook tutorial demonstrating point cloud Hausdorff distances.
• Added dedicated bash orchestration scripts (fairodt.sh, fairvol.sh, etc.) for bulk DMTet evaluations.