zhangyuanyang21 / 2023-cvpr-gcfaggmvc Goto Github PK

Authors: Weiqing Yan, Yuanyang Zhang, Chenlei Lv, Chang Tang*, Guanghui Yue, Liang Liao, Weisi Lin.

This repo contains the code and data of our CVPR'2023 paper GCFAgg: Global and Cross-View Feature Aggregation for Multi-View Clustering.

The overall framework. Our module includes global and cross-view feature aggregation module (GCFAgg) and structure-guided multiview contrastive learning module (SgCL). The former learns a consensus representation via considering global structure relationship among samples, which fully explores the complementary of similar samples. The latter integrates the learnt global structure relationship and consensus representation to contrastive learning, which makes data representations in the same cluster similar and addresses the aforementioned second issue in the introduction. Note that, EC: Encoder; DC: Decoder; Cat:Concatenation; MLP: Multi-Layer Perception.

pytorch==1.12.1

numpy>=1.21.6

scikit-learn>=1.0.2

The Synthetic3d, Prokaryotic, and MNIST-USPS datasets are placed in "data" folder. The others dataset could be downloaded from cloud. key: data

The code includes:

• an example for train a new model：

python train.py

• an example for test the trained model:

python test.py

You can get the following output:

Epoch 290 Loss:15.420288
Epoch 291 Loss:15.431067
Epoch 292 Loss:15.417261
Epoch 293 Loss:15.436375
Epoch 294 Loss:15.398655
Epoch 295 Loss:15.406467
Epoch 296 Loss:15.413018
Epoch 297 Loss:15.419146
Epoch 298 Loss:15.419894
Epoch 299 Loss:15.389602
Epoch 300 Loss:15.377309
---------train over---------
Clustering results:
ACC = 0.9700 NMI = 0.8713 PUR=0.9700 ARI = 0.9126
Saving model...

How to obtain S:

If you can't obtain S you can refer Obtain-S.docx

Work&Code is inspired by MFLVC, CONAN, CoMVC ...

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

@InProceedings{Yan_2023_CVPR,
    author    = {Yan, Weiqing and Zhang, Yuanyang and Lv, Chenlei and Tang, Chang and Yue, Guanghui and Liao, Liang and Lin, Weisi},
    title     = {GCFAgg: Global and Cross-View Feature Aggregation for Multi-View Clustering},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {19863-19872}
}

If you have any problems, contact me via [email protected].