Release v0.3.2
Features:
- Added BMesh batched mesh data structure and batched mesh collate functions.
- Implemented mesh to sparse tensor conversion logic.
- Implemented core transformations, vertex operations, and dataset transformations.
- Updated Digit3D dataset implementation.
Refactors & Build:
- Refactored conquer3d and data_structure module initialization.
- Fixed sparse2voxel off-by-one vertex coordinate bug.
- Removed outdated API documentation, examples, and mkdocs configuration.
- Updated project configuration.
Examples:
- Re-architected Sparse VAE for improved encoding.
- Fixed TorchSparse border coordinate dropping bug in the VAE via dense cache mapping.
- Added AdamW optimizer, learning rate scheduler, and dropout regularization to the Sparse Classifier.
- Integrated Digit3D into Sparse Classification.
- Updated and fixed inference notebooks for Classification and Reconstruction.
- Added test history and visualizer notebooks for Sparse VAE architecture.