CVPR 2025
PaCo implements parametric completion, a new point cloud completion paradigm that recovers parametric primitives rather than individual points, for polygonal surface reconstruction.
Simply click the badge below to run the demo:
Before you begin, ensure that your system has the following prerequisites installed:
- Conda
- CUDA Toolkit
- gcc & g++
The code has been tested with Python 3.10, PyTorch 2.6.0 and CUDA 11.8.
-
Clone the repository and enter the project directory:
git clone https://github.com/parametric-completion/paco && cd paco
-
Install dependencies:
Create a conda environment with all required dependencies:
. install.sh
-
Download the preprocessed ABC data:
to
./data/abc:python ./scripts/download_data.py
-
(Optional) Download pretrained weights:
to
./ckpt/ckpt-best.pth:python ./scripts/download_ckpt.py
-
Start training using one of the two parallelization:
Distributed Data Parallel (DDP):
# Replace device IDs with your own CUDA_VISIBLE_DEVICES=0,1 ./scripts/train_ddp.shData Parallel (DP):
# Replace device IDs with your own CUDA_VISIBLE_DEVICES=0,1 ./scripts/train_dp.sh -
Monitor training progress using TensorBoard:
# Replace ${exp_name} with your experiment name (e.g., default) # Board typically available at http://localhost:6006 tensorboard --logdir './output/${exp_name}/tensorboard'
-
Start evaluation of the reconstruction:
# Default checkpoint at `./ckpt/ckpt-best.pth` CUDA_VISIBLE_DEVICES=0,1 ./scripts/test.shThe results will be saved to
${output_dir}/evaluation.csv.
# Check available configurations for training
python train.py --cfg job
# Check available configurations for evaluation
python test.py --cfg jobAlternatively, review the main configuration file: conf/config.yaml.
- Demo and pretrained weights
- Dataset and evaluation script
- Hugging Face model
If you use PaCo in a scientific work, please consider citing the paper:
[paper] [supplemental] [arxiv] [bibtex]
@InProceedings{chen2025paco,
title = {Parametric Point Cloud Completion for Polygonal Surface Reconstruction},
author = {Zhaiyu Chen and Yuqing Wang and Liangliang Nan and Xiao Xiang Zhu},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2025},
pages = {11749-11758}
}Part of our implementation is based on the PoinTr repository. We thank the authors for open-sourcing their great work.