Release v0.3.1

Features

• Added Sparse VAE (SimpleSparseVAE) with coarse-to-fine generative architecture and reconstruction loop.
• Added SparseClassifier model architecture and updated training loop with automated checkpointing.
• Implemented SparseDigit3D and Digit3D native dataset loaders with automated sparse_collate_fn.
• Added sparse2mesh_topology C++ bindings and Python wrapper for grid topology extraction.
• Developed SDF computation and mesh construction scripts for Digit3D dataset generation.

Refactors & Build

• Added torchsparse and open3d to the project dependencies.
• Refactored AABB primitives for integration with grid topology extraction.
• Updated .gitignore for local model checkpoints and raw data directories.
• Enhanced README documentation detailing new vision APIs and dataset functionalities.

Examples

• Created sparse_reconstruction_infer.ipynb for Sparse VAE topological interpolation and mesh reconstruction visualizations.
• Created sparse_classification_infer.ipynb to visualize classification inferences alongside matching reconstructed meshes.
• Added a visualization notebook for Digit3D data exploration.
• Updated marching_cubes.ipynb and sparse_tensor.ipynb examples for spatial data structures.