RePart

Project Page

Setup

Use Python 3.10 or newer:

pip install -r requirements.txt

Provide a mesh directory with UID-named files (for example, /path/to/meshes/<uid>.obj) and a matching PartNet point-sample directory:

/path/to/partnet_points/
`-- <uid>/
    `-- point_sample/
        |-- label-10000.txt
        |-- ply-10000.ply
        `-- pts-10000.txt

The mesh filename stem must match the UID directory. Meshes can be OBJ, PLY, STL, OFF, GLB, or GLTF. To use precomputed SDF CSV files, pass --sdf-root to the training script; otherwise SDFs are generated from the input meshes.

Train

bash scripts/train_category.sh --data-root /path/to/meshes --point-sample-root /path/to/partnet_points --out-dir /path/to/category_run --rollout-episodes-per-epoch 100
bash scripts/train_joint.sh --run-dir /path/to/little_run --run-dir /path/to/container_run --run-dir /path/to/furniture_run --out-dir /path/to/joint_run

Evaluate

bash scripts/evaluate.sh --checkpoint /path/to/joint_run/sq_partnet_rl_epoch_0100.pt --data-root /path/to/test_meshes --point-sample-root /path/to/partnet_points --out-dir /path/to/eval_run

The output directory contains grouped SQs and projected mesh labels.

Tests

python -m unittest discover -s tests