Workflow status
This repository contains research code with several levels of support.
Supported for public sanity checks
- Hydra config composition for every runner-backed paper recipe.
- CLI rendering of reproduction commands through
scripts/reproduce_experiment.py.
- CPU-safe unit tests for utilities, datasets, evaluation helpers, and public APIs.
- Source-only linting and type checking through the validation scripts.
Requires local data or checkpoints
- Paper reproduction beyond command/config sanity.
gen_eval, mesh_eval, and BOP scene evaluation on real datasets.- Any recipe that accepts
weights, vae_weights, occ_weights, or inst_weights.
Requires compiled or GPU dependencies
- CUDA extension tests for Chamfer/EMD/fusion/MISE-style components.
- tiny-cuda-nn backends.
- GPU training and full evaluation runs.
- Renderer tests on systems without stable headless rendering support.
Experimental
- Long-running sweeps and ablation scripts.
- Some visualization and rendering workflows.
- Optional third-party backends that are not installed by the base package.
When adding public workflows, prefer a CPU-safe sanity test first and keep machine-specific setup in local, untracked configuration.