Introduction

This repo provides the implementation of ClusMatch: Improving Deep Clustering by Unified Positive and Negative Pseudo-label Learning.

Framework

Main Results

Training

Preparing Environment

use pip to install required packages:

pip install -r requirements.txt

Preparing Pre-trained Model

Our method requires pre-trained models. Therefore, you need to download the checkpoints from the 'pretrain' path and specify the 'load_path' in the configs.

Additionally, there are also some training logs available. Google Drive

Training Commands

# ProPos+CIFAR-10+ResNet-18
python train.py --c config/classic_cv/clusmatch/clusmatch_cifar10_600_0_propos.yaml

# ProPos+CIFAR-100+ResNet-18
python train.py --c config/classic_cv/clusmatch/clusmatch_cifar100_6000_0_propos.yaml

Please check if the "load_path" is correct before training.