dushide / umcgl_tip_2024

This is an implement of methods with PyTorch in "UMCGL: Universal Multi-view Consensus Graph Learning with Consistency and Diversity" that published in IEEE Transactions on Image Processing.

Universal Multi-view Consensus Graph Learning (UMCGL)

Introduction

This is an implement of methods with PyTorch in "UMCGL: Universal Multi-view Consensus Graph Learning with Consistency and Diversity" that published in IEEE Transactions on Image Processing. UMCGL is run on a server with a standard Ubuntu-16.04 operation system with a NVIDIA Tesla P100 16G GPU and 126G RAM.

Datasets Descriptions

• Original multi-view datasets are stored in /_multiview datasets.
• Original graph matrices are stored in /datasetW, which are generated by KNN (see Matlab codes in /datasetW/ConstructW files).
• For all datasets, please obtain them from the following links: https://drive.google.com/drive/folders/1Po0Im25dogk4scK0Sm_FqaGg1P93F4Hd?usp=drive_link; https://pan.baidu.com/s/1bUUz3lEWTi1V8pEemTh0Aw?pwd=59e0 (Extracted code：59e0).

Requirements

Require Python 3.7.2

• torch 1.7.1
• numpy 1.21.6
• scikit-learn 1.0.2

Quick Running and Testing

• Run run_example.py for multi-view complete and incomplete clustering tasks to see the example performance.
• If you want to test your own datesets, you can put the datasets into /_multiview datasets and obtain their graphs by ConstructW.zip. Then, you can move them into /datasetW and train the corresponding models by employing these graphs. Finally, you can set the configuration file according to the example /config.

Reference

@ARTICLE{Du2024UMCGL,
  author={Du, Shide and Cai, Zhiling and Wu, Zhihao and Pi, Yueyang and Wang, Shiping},
  journal={IEEE Transactions on Image Processing}, 
  title={{UMCGL}: Universal Multi-view Consensus Graph Learning with Consistency and Diversity}, 
  year={2024},
  volume={33},
  pages={3399-3412},
}

If you have any questions, please feel free to contact [email protected] at any time. Thanks.