The official implementation of our NeurIPS 2024 paper "Unlearnable 3D Point Clouds: Class-wise Transformation Is All You Need", by Xianlong Wang, Minghui Li, Wei Liu, Hangtao Zhang, Shengshan Hu, Yechao Zhang, Ziqi Zhou, and Hai Jin.
Traditional unlearnable strategies have been proposed to prevent unauthorized users from training on the 2D image data. With more 3D point cloud data containing sensitivity information, unauthorized usage of this new type data has also become a serious concern. To address this, we propose the first integral unlearnable framework for 3D point clouds including two processes: (i) we propose an unlearnable data protection scheme, involving a class-wise setting established by a category-adaptive allocation strategy and multi-transformations assigned to samples; (ii) we propose a data restoration scheme that utilizes class-wise inverse matrix transformation, thus enabling authorized-only training for unlearnable data. This restoration process is a practical issue overlooked in most existing unlearnable literature, i.e., even authorized users struggle to gain knowledge from 3D unlearnable data. Both theoretical and empirical results (including 6 datasets, 16 models, and 2 tasks) demonstrate the effectiveness of our proposed unlearnable framework.
| Date | Event |
|---|---|
| 2024/10/21 | We have released the official implementation of UnlearnablePC! |
| 2024/10/07 | The paper (arXiv version) is available at UnlearnablePC! |
| 2024/09/26 | UnlearnablePC is acccepted by NeurIPS 2024! |
- Get code
git clone https://github.com/CGCL-codes/UnlearnablePC.git- Build environment
cd UnlearnablePC
conda create -n UnlearnablePC python=3.9
conda activate UnlearnablePC
pip install -r requirements.txt-
Download datasets
- Please download ModelNet dataset at: [ModelNet], ShapeNetPart dataset at: [ShapeNetPart], ScanObjectNN dataset at: [ScanObjectNN], S3DIS dataset at: [S3DIS] (for semantic segmentation task)
- Unzip the datasets.zip files in
UnlearnablePC/clean_data
-
Produce unlearnable 3D point cloud datasets
-
Using UMT (k=4) with default settings and parameters (robust unlearnable dataset)
python UnlearnablePC_generation.py --dataset ModelNet10 --mode "['rot', 'scale', 'twist', 'shear']"- Using UMT (k=3) with default settings and parameters (robust unlearnable dataset)
python UnlearnablePC_generation.py --dataset ModelNet10 --mode "['rot', 'scale', 'twist']"- Using UMT (k=2) with default settings and parameters (fragile to SE(3)-invariance networks)
python UnlearnablePC_generation.py --dataset ModelNet10 --mode "['rot', 'scale']"- Training on UMT datasets (i.e., unlearnable datasets)
python train.py --dataset ModelNet10 --target_model pointnet_cls --UMTK 3- Training on clean datasets (i.e., raw 3D point cloud datasets)
python train.py --dataset ModelNet10 --target_model pointnet_cls --clean_train- Employing the data restoration scheme to UMT data (for authorized users)
python reverse_train.py --dataset ModelNet10 --target_model pointnet_cls --mode "['rot', 'scale', 'twist']"If you find UnlearnablePC both interesting and helpful, please consider citing us in your research or publications:
@inproceedings{wang2024unlearnable,
title={Unlearnable 3D Point Clouds: Class-wise Transformation Is All You Need},
author={Wang, Xianlong and Li, Minghui and Liu, Wei and Zhang, Hangtao and Hu, Shengshan and Zhang, Yechao and Zhou, Ziqi and Jin, Hai},
booktitle={Proceedings of the 38th Annual Conference on Neural Information Processing Systems (NeurIPS'24)},
year={2024}
}