/SADH

Self-supervised asymmetric deep hashing with margin-scalable constraint for image retrieval

Primary LanguagePython

Deep-hashing-with-self-supervised-asymmetric-semantic-excavation-and-margin-scalable-constraint

Pytorch implementation of our paper for deep hashing retrieval.

Deep hashing with self-supervised asymmetric semantic excavation and margin-scalable constraint

by Zhengyang Yu, Song Wu*, Zhihao Dou and Erwin M.Bakker

Neurocomputing, 2022

Introduction

This repository is Pytorch implementation of SADH, which mainly deals with deep hashing retrieval under multi-label scenario. The main insights of SADH are: 1) an asymmetric semantic learning strategy and 2) a margin-scalable similarity constraint. The network structure is illustrated as follows: avatar

How to use

You can easily train and test SADH via running:

  labnet.py
  
  imgnet.py

You can download the datasets via:

NUSWIDE

MS-COCO 2014

MIR-Flickr25k

Cite this paper

  @article{article,
  title = {Deep hashing with self-supervised asymmetric semantic excavation and margin-scalable constraint},
  journal = {Neurocomputing},
  volume = {483},
  pages = {87-104},
  year = {2022},
  doi = {https://doi.org/10.1016/j.neucom.2022.01.082},
  author = {Zhengyang Yu and Song Wu and Zhihao Dou and Erwin M. Bakker},
  keywords = {Deep supervised hashing, Asymmetric learning, Self-supervised learning}
  }

Acknowledgement

Thanks for the work of swuxyj. Our code is heavily borrowed from the implementation of [https://github.com/swuxyj/DeepHash-pytorch].