Code for Mitigating Object Dependencies: Improving Point Cloud Self-Supervised Learning through Object Exchange (OESSL) CVPR2024
-----------------------------------------------------------------------------------
Todo list:
1、Now the grapcut based unsupervised segmentation is not included in this project, since it is writen using C++. I will update it into this project as soon as possiable.
2、The fine-tune code will be updated to this project
3、I will continuously optimize this project.
Our project is built based on STSSL
Installing pre-requisites:
sudo apt install build-essential python3-dev libopenblas-dev
pip3 install -r requirements.txt
pip3 install torch ninja
Installing MinkowskiEngine with CUDA support:
pip3 install -U MinkowskiEngine==0.5.4 --install-option="--blas=openblas" -v --no-deps
1、Download ScanNet 2、Next, preprocess all scannet raw point cloud follwing "https://github.com/chrischoy/SpatioTemporalSegmentation" 3、Segment the pointclouds and generate box follwing "https://github.com/chrischoy/SpatioTemporalSegmentation/blob/master/README.md" or you can download the segments and boxes here.
for pre-training. (We use 8 RXT3090 GPUs for pre-training)
you can just run train.py remember to modify the paramters of path : )
Then for fine-tuning:
You can refer to SpatioTemporalSegmentation
Any questions, touch me at wuyanhao@stu.xjtu.edu.cn
You can download the pre-trained model on ScanNet here: https://drive.google.com/file/d/1NOofbACbj79WqVSzdKqdrOcBtUPwkfHK/view?usp=drive_link
If you use this repo, please cite as :
@inproceedings{wu2024mitigating,
title={Mitigating Object Dependencies: Improving Point Cloud Self-Supervised Learning through Object Exchange},
author={Wu, Yanhao and Zhang, Tong and Ke, Wei and Qiu, Congpei and S{\"u}sstrunk, Sabine and Salzmann, Mathieu},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={23052--23061},
year={2024}
}