Awesome

PyTorch implementation for

Semantic Invariant Multi-view Clustering with Fully Incomplete Information

TPAMI 2023

[Paper] [Discussion] [More Information]

Introduction

SMILE framework

Requirements

• Python 3.10
• PyTorch 1.12.1
• faiss

conda install -c pytorch faiss-gpu

Training

Modify the ./Utils/PathPresettingOperator.get_dataset_path, then train the model(s):

# NoisyMNIST 
python main.py --dataset NoisyMNIST30000 --seed 9116  --aligned_prop 1 --complete_prop 1
  
# MNISTUSPS 
python main.py --dataset MNISTUSPS --seed 9116  --aligned_prop 1 --complete_prop 1
  
# Caltech     
python main.py --dataset 2view-caltech101-8677sample --seed 9116    --aligned_prop 1 --complete_prop 1
  
# CUB 
python main.py --dataset cub_googlenet_doc2vec_c10 --seed 9116    --aligned_prop 1 --complete_prop 1

# YouTubeFaces    
python main.py --dataset YouTubeFaces --seed 9116  --aligned_prop 1 --complete_prop 1

Model Zoo

The pre-trained models are available here.

Download the models, then:

python main.py --dataset dataset --seed seed --resume PathToYourModel

Experiment Results:

Citation

If SMILE is useful for your research, please cite the following paper:

@article{zeng2023semantic,
  title={Semantic Invariant Multi-view Clustering with Fully Incomplete Information},
  author={Zeng, Pengxin and Yang, Mouxing and Lu, Yiding and Zhang, Changqing and Hu, Peng and Peng, Xi},
  journal={arXiv preprint arXiv:2305.12743},
  year={2023}
}