2024-MM-VITAL

[ACM MM 2024] Code for the paper "Robust Variational Contrastive Learning for Partially View-unaligned Clustering]"

Important

🎉 New Update: The codebase now supports more than 2 views!

• ✅ Complete multi-view clustering with arbitrary number of views
• ✅ Partially View-unaligned (PVP) setting with arbitrary number of views
• ⚡ Faster training with PyTorch-based GMM implementation (replacing sklearn's numpy-based GMM)

---

Quick Start

1. Prepare Your Dataset

Modify the load_data() function in run-multi-view-version.py to load your own dataset:

elif args.dataset_name == 'YourDataset':
    # Load each view into data_list
    data_list.append(mat['view1'])  # View 1: shape (n_samples, dim1)
    data_list.append(mat['view2'])  # View 2: shape (n_samples, dim2)
    data_list.append(mat['view3'])  # View 3: shape (n_samples, dim3) (optional)
    # ... add more views as needed
    label = np.squeeze(mat['label'].astype(np.uint8))

2. Configure Parameters

Key parameters in run-multi-view-version.py:

Parameter	Description	Default
--dataset_name	Name of your dataset	'Scene15'
--dataset_path	Path to dataset folder	'./datasets/'
--aligned_rate	Alignment ratio (1.0 = fully aligned, 0.5 = 50% PVP)	1.0
--train_time	Number of training runs for averaging	5

3. Run Training

python run-multi-view-version.py --dataset_name 'YourDataset' --aligned_rate 0.5 --train_time 5

---

Experiment Results

Partially Aligned (50%) Clustering Performance

Fully Aligned (100%) Clustering Performance

---

Citation

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

@inproceedings{he2024robust,
  title={Robust Variational Contrastive Learning for Partially View-unaligned Clustering},
  author={He, Changhao and Zhu, Hongyuan and Hu, Peng and Peng, Xi},
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},
  pages={4167--4176},
  year={2024}
}