PPCC: Syn-to-Real Domain Adaptation for Point Cloud Completion via Part-based Approach (ECCV 2024)

Official code repository for the ECCV 2024 paper:
"Syn-to-Real Domain Adaptation for Point Cloud Completion via Part-based Approach"
by Yunseo Yang, Jihun Kim, and Kuk-Jin Yoon.

Setup

We recommend using conda to create a virtual environment.

conda create -n ppcc python=3.8
conda activate ppcc
pip install -r requirements.txt

Install

To enable CUDA-based acceleration, install the required extensions:

cd extensions/chamfer_dist/ && python setup.py install
cd extensions/emd/ && python setup.py install
cd utils/pointnet2_ops_lib/ && python setup.py install

Datasets

The preprocessed datasets are available here.

• PartNet: A synthetic point cloud dataset with per-point part labels.
• USSPA: A real-world point cloud dataset with per-point part labels and confidence values.

The original USSPA dataset can be found at USSPA.
We reprocessed the data by adding part labels and confidence values for use in our training and evaluation pipelines.

Please organize the dataset directory as follows:

data/
├── chair/
│   └── partnet
│       └── train
│       └── val
│       └── test
│   └── usspa
│       └── ...
├── table/
│   └── ...
└── ...

Make sure to update dataset paths in configs/PPCC.yaml if needed.

Train

To train the model, run the following command:

CUDA_VISIBLE_DEVICES={GPU_ID} python main.py --category {category} --exp_name {name}

• --category: Object category to train on. Options are: chair, table, lamp, bed
• --exp_name: A custom name for the experiment (used for logging and checkpoints).

Example:

CUDA_VISIBLE_DEVICES=0 python main.py --category chair --exp_name chair_exp1

Evalutation

To evaluate the model, use the following command:

CUDA_VISIBLE_DEVICES={GPU_ID} python main.py --evaluation --category {category} --ckpts {checkpoint_path}

• --evaluation: Flag to indicate evaluation mode.
• --category: Object category to evaluate on. Options are: chair, table, lamp, bed
• --ckpts: Path to the saved model checkpoint for evaluation.

Citation

If you find our work useful for your research, please consider citing:

@inproceedings{yang2024syn,
  title={Syn-to-Real Domain Adaptation for Point Cloud Completion via Part-Based Approach},
  author={Yang, Yunseo and Kim, Jihun and Yoon, Kuk-Jin},
  booktitle={European Conference on Computer Vision},
  pages={179--197},
  year={2024},
  organization={Springer}
}