PointConv: Deep Convolutional Networks on 3D Point Clouds. CVPR 2019
Wenxuan Wu, Zhongang Qi, Li Fuxin.
This project is based on repo PointConv. I tried to reproduct the paper using code in that repo. But I found that the train consumes almost all of my RAM, leading to out of memory. So I optimized the data augmentation steps to be performed on the GPU and fixed a bug caused by scikit-learn update.
@inproceedings{wu2019pointconv,
title={Pointconv: Deep convolutional networks on 3d point clouds},
author={Wu, Wenxuan and Qi, Zhongang and Fuxin, Li},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={9621--9630},
year={2019}
}
The code is modified from repo PointConv. Please install PyTorch, pandas, and sklearn. The code has been tested with Python 3.12, pytorch 2.6, CUDA 12.6 on Windows 11.
Download the ModelNet40 dataset from here. This dataset is the same one used in PointNet, thanks to Charles Qi. Copy the unziped dataset to ./data/modelnet40_normal_resampled.
To train the model,
python train_cls_conv.py --model pointconv_modelnet40 --normal
To evaluate the model,
python eval_cls_conv.py --checkpoint ./checkpoints/checkpoint.pth --normal
This repository is released under MIT License (see LICENSE file for details).