EIP

Official PyTorch implementation for the following paper:

Incorporating Rotation Invariance with Non-invariant Networks for Point Clouds

by Jiajun Fei, Zhidong Deng

TL;DR: We propose a novel rotation invariant representation learning method for point clouds named EIPs, which bridge the gap between the non-invariant point cloud representation learning and the invariant one. With EIPs, well-performed non-invariant networks can be easily deployed in invariant tasks and applications.

OpenPoints

The repository is built upon OpenPoints, which is a library built for fairly benchmarking and easily reproducing point-based methods for point cloud understanding. It is born in the course of PointNeXt project and is used as an engine therein.

For any question related to OpenPoints, please open an issue in PointNeXt repo.

Citation

@Article{qian2022pointnext,
  author  = {Qian, Guocheng and Li, Yuchen and Peng, Houwen and Mai, Jinjie and Hammoud, Hasan and Elhoseiny, Mohamed and Ghanem, Bernard},
  title   = {PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies},
  journal = {arXiv:2206.04670},
  year    = {2022},
}

We modify OpenPoints in the following aspects:

• Implement EIP under openpoints/models/pose.
• Add validate_so3 support in examples/classification/train.py and examples/shapenetpart/main.py.
• Remove most unrelated files and packages.
• Make some other slight modifications.

Requirements

conda create -n eip -y python=3.7
conda activate eip
conda install -y pytorch=1.10.1 torchvision cudatoolkit=11.3 -c pytorch -c nvidia
pip install -r requirements.txt
cd openpoints/cpp/pointnet2_batch
python setup.py install

Datasets

Fill in the datasets path of all cfgs/DATASET/default.yaml. OpenPoints support automatic downloading and processing.

Train & Test

Point Cloud Classification

# train
CUDA_VISIBLE_DEVICES=$GPUs python examples/classification/main.py --cfg $cfg

# test so(3)
CUDA_VISIBLE_DEVICES=$GPUs python examples/classification/main.py --cfg $cfg --pretrained_path $ckpt_path --mode test_so3 --so3_mode random_rot --num_rot 1

Point Cloud Part Segmentation

# train
CUDA_VISIBLE_DEVICES=$GPUs python examples/shapenetpart/main.py --cfg $cfg

# test so(3)
CUDA_VISIBLE_DEVICES=$GPUs python examples/shapenetpart/main.py --cfg $cfg --pretrained_path $ckpt_path --mode test_so3 --so3_mode random_rot --num_rot 1

Release checkpoints

We release all EIP checkpoints with log files. You can get our results by unzipping the downloaded file under release_ckpt/ and running evaluate.sh. Download links: [百度网盘]、[Google Drive].

Acknowledgements

Our codes are built upon OpenPoints. Thanks to their excellent works!

Reference

placeholder