One-pass Multi-view Clustering for Large-scale Data (OPMC)

An unofficial pytorch-based implementation of the paper:

Jiyuan Liu, Xinwang Liu, Yuexiang Yang, Li Liu, Siqi Wang, Weixuan Liang and Jiangyong Shi: One-pass Multi-view Clustering for Large-scale Data, IEEE International Conference on Computer Vision, ICCV, 2021. (Accepted Jul. 2021)

The original MATLAB-based implementation could be found here: https://github.com/liujiyuan13/OPMC-code_release

Why is it useful?

• Non-deep, parameter-free multiview clustering algorithm
• Comparative results to the recent baselines
• Could be used with CPU only

How to run?

1. Unzip a dataset from datasets directory or prepare your own dataset
2. Run the next command: python run.py --dataset <PATH TO PICKLED DATASET>

The dataset is assumed to be stored as a dictionary with keys:

• X - dict with key, value for each view (key = name of the view, value = numpy ndarray matrix for samples)
• Y - dict with key, value for each view (key = name of the view, value = numpy ndarray matrix for labels )

Example datasets:

• MSRCv1 Consists of 210 scene recognition images belonging to 7 categories Zhao et al. (2020). Each image is described by 5 different types of features.
• Scene15 Consists of 4,485 scene images belonging to 15 classes Fei-Fei & Perona (2005).
• RBGD Kong et al. (2014): Consists of 1, 449 samples of indoor scenes image-text of 13 classes. We follow the version provided in Trosten et al. (2021a); Zhou & Shen (2020), where image features are extracted from a ResNet50 model pretrained on the ImageNet dataset and text features from a doc2vec model pretrained on the Wikipedia dataset