Awesome

🍬 CANDY: Contextually-spectral based correspondence refinery

This repository contains the PyTorch implementation for CANDY proposed in paper Robust Contrastive Multi-view Clustering against Dual Noisy Correspondence (NeurIPS 2024).

Requirements

Environment

The code depends on pytorch, scikit-learn, scipy and munkres. You can set up your environment with the following commands:

conda create -n candy python=3.11 pytorch==2.2.2 pytorch-cuda=11.8 -c pytorch -c nvidia -y
conda activate candy
pip install scikit-learn==1.3.0 scipy==1.12.0 munkres==1.1.4

Datasets

Dataset used in the paper can be downloaded from Google Drive or Baidu NetDisk. We assume that you have placed the downloaded datasets in the directory /path/to/your/data.

Running

Please replace /path/to/your/data with the directory of your downloaded datasets. Note that arguments in the --config_file overrides other command line arguments.

EXPERIMENT_ROOT=/path/to/your/data python main_train.py --config_file config/Scene15.yaml --fp_ratio 0.5 --epochs 200

Reference

If you find our work useful in your research, please consider citing:

@article{guo2024candy
    title = {Robust Contrastive Multi-view Clustering against Dual Noisy Correspondence},
    author = {Ruiming Guo and Mouxing Yang and Yijie Lin and Xi Peng and Peng Hu},
    booktitle = {Advances in Neural Information Processing Systems},
    volume = {37},
    year = {2024}
}

Acknowledgement

This implementation is based on DIVIDE.