r0f1 / distribution_clustering

Implements (high-dimenstional) clustering algorithm described in https://arxiv.org/pdf/1804.02624.pdf

Description

Implements (high-dimenstional) clustering algorithm described in https://arxiv.org/pdf/1804.02624.pdf

Dependencies

python3
pytorch (>=0.4)
torchvision
PILLOW
numpy
scipy
tqdm

Usage

First, if you want to use the default BaseDataset class, the directory structure of the data you wish to be clustered must conform to the structure shown below. If another structure makes more sense for your purposes, you will need to sublass the BaseDataset class and reference your class in save_dataset_features.py and cluster_dataset.py.

- <data_dir>
    - samples
        - <first file>
        - ...

You can then extract deep features from your data by running the command below. If you are using the BaseDataset class, your features will be saved at the path <data_dir>/features.npy

python sample_dataset_features.py --data_dir <path to data directory>

Finally, you can cluster your data by running the command below. If you are using the BaseDataset class, your clustered data will be saved at the path <data_dir>/clusters. Parameters within brackets () are optional.

python cluster_dataset.py --data_dir <path to data directory> (--thres <float>) (--min_clus <int>) (--max_dist <float>) (--dont_normalize)