Hierarchical Quantization Consistency for Fully Unsupervised Image Retrieval

Guile Wu (Researcher), Chao Zhang (Toshiba Europe Limited),* Stephan Liwicki (Toshiba Europe Limited)
The 34th British Machine Vision Conference


Unsupervised image retrieval aims to learn an efficient retrieval system without ex- pensive data annotations. Typical methods rely on handcrafted feature descriptors or pre-trained feature extractors. Recent advances propose deep fully unsupervised image retrieval aiming at training a deep model from scratch to jointly optimize visual features and quantization codes with minimal human supervision. Thus approach mainly focuses on instance contrastive learning without using semantic information. However, a fun- damental problem of contrastive learning is mitigating the effects of false negatives. To this end, we exploit sub-quantized representations to extract fine-grained semantics for self-supervised learning. To further regularize the instance contrastive learning for quan- tization, we also leverage consistency regularization to reflect the similarities between the query sample and negative samples. Specifically, we propose a novel hierarchical consistent quantization approach to deep fully unsupervised image retrieval, which con- sists of part consistent quantization and global consistent quantization. With a unified learning objective, our approach exploits richer self-supervision cues to facilitate model learning. Extensive experiments on three benchmark datasets show the superiority of our approach over the state-of-the-art methods.



author    = {Guile Wu and Chao Zhang and Stephan Liwicki},
title     = {Hierarchical Quantization Consistency for Fully Unsupervised Image Retrieval},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {https://papers.bmvc2023.org/0215.pdf}

Copyright © 2023 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection