v0.6.3: Coarse-to-Fine Hierarchical Volumetric Flood Fill

Release Notes - Conquer3D v0.6.3

Features

• Unified 2-Level Coarse-to-Fine (CF) Volumetric Flood Fill (sign_mode=5):
  • Implemented high-performance hierarchical flood fill engine reducing GPU memory by $314.7\times$ at $1024^3$ resolution ($9.2\text{ GB} \to 29.28\text{ MB}$).
  • Designed a Unified Multi-Scale Graph Wavefront Propagation kernel allowing water to flow seamlessly across empty coarse macro-blocks and fine boundary shared-memory blocks.
  • Added exact 3D Separating Axis Theorem (SAT) triangle-box intersection tests (Triangle::is_voxel_intersect) and hardware-accelerated BVH ray-segment tests for all inter-scale transitions.
  • Achieved $100.000%$ exact topological parity and $0$ open boundary edges across Marching Cubes, Dual Contouring, Dual Marching Cubes, and Marching Tetrahedra.
• MeshBVH & TriangleMesh Sign Mode 5:
  • Integrated sign_mode=5 into MeshBVH.query_points() and TriangleMesh.query_points() with $O(1)$ early-exit on coarse water/interior blocks.
  • Added TriangleMesh.build_flood_fill_cf_data() with automatic dynamic macro-block divisor selection.
• Conversion Pipelines Integration:
  • Wired sign_mode=5 support into conquer3d.conversion.tmesh2voxel and conquer3d.conversion.tmesh2sparse.
• Benchmark Models Expansion:
  • Added standard Stanford 3D models catalog loaders (conquer3d.data.assets.Armadillo, Bunny, Spot, etc.).

Refactors & Build

• Dynamic Symbol Resolution: Enabled RTLD_GLOBAL import flag for robust C++ extension symbol linking.
• PyPI & C++ Bindings: Registered standalone flood_fill_cf submodule and generated comprehensive .pyi type stubs.
• Version Bump: Bumped version to 0.6.3 in pyproject.toml.

Examples

• Standalone verification and benchmark scripts comparing Dense vs Coarse-to-Fine flood fill across $256^3, 512^3, 1024^3$ resolutions with watertight Marching Cubes, Dual Contouring, and Dual Marching Cubes extraction.