GlocalMatch for Open-Domain SSL

Introduction

This repository hosts the code for our paper accepted by IEEE Transactions on Knowledge and Data Engineering (TKDE):

Open-Domain Semi-Supervised Learning via Glocal Cluster Structure Exploitation
Zekun Li, Lei Qi, Yawen Li, Yinghuan Shi*, Yang Gao

[Preprint][BibTeX]

Preparation

Required Packages

We suggest first creating a conda environment:

conda create --name glocalmatch python=3.8

then use pip to install required packages:

pip install -r requirements.txt

Datasets

Please download the data files from this link and put them into the ./data folder:

GlocalMatch
├── config
    └── ...
├── data
    ├── cifarstl
        └── cifar
        └── stl
    └── domainnet_balanced
        └── ...
    └── pacs
        └── ...
├── semilearn
    └── ...
└── ...  

Usage

We implement GlocalMatch using the codebase of USB.

Here is an example to train GlocalMatch on the CIFAR-STL benchmark, with CIFAR as the labeled domain, and 45 labels available per class.

# seed = 1
CUDA_VISIBLE_DEVICES=0 python train.py --c config/opendomain_cv/glocalmatch/glocalmatch_cifarstl_c45_1.yaml

For other tasks, the config files have been released in ./config/opendomain_cv.

Citation

@article{glocalmatch,
  title={Open-Domain Semi-Supervised Learning via Glocal Cluster Structure Exploitation},
  author={Li, Zekun and Qi, Lei and Li, Yawen and Shi, Yinghuan and Gao, Yang},
  journal={IEEE TKDE},
  year={2024}
}