Skip to content
/ OESSL Public

Code for **Mitigating Object Dependencies: Improving Point Cloud Self-Supervised Learning through Object Exchange** (OESSL) CVPR2024

Notifications You must be signed in to change notification settings

YanhaoWu/OESSL

Repository files navigation

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.

Paper | Project page

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

Data Preparation

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.

Reproducing the results

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 [email protected]

You can download the pre-trained model on ScanNet here: https://drive.google.com/file/d/1NOofbACbj79WqVSzdKqdrOcBtUPwkfHK/view?usp=drive_link

Citation

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}
}

About

Code for **Mitigating Object Dependencies: Improving Point Cloud Self-Supervised Learning through Object Exchange** (OESSL) CVPR2024

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages