This repository contains an implementation to the ECCV 2020 paper: "PointMixup: Augmentation for point cloud".
- Required packages: python3.6, pytorch 1.2, tensorboardX
- pip install git+git://github.com/erikwijmans/etw_pytorch_utils.git@v1.1.1#egg=etw_pytorch_utils
- pip install --no-cache --upgrade git+https://github.com/dongzhuoyao/pytorchgo.git
- sh setup.sh
Baseline:
python main.py -savename exp0 -pointmixup False -manimixup False
Our method (point mixup):
python main.py -savename exp1 -pointmixup True -manimixup False -mixup_alpha 0.4
Our method (point mixup + manifold mixup):
python main.py -savename exp2 -pointmixup True -manimixup True -mixup_alpha 1.5
If data cannot be downloaded automatically from the script, please manually download from 'https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip' and extract to './pointnet2/data/' folder.