Self-adaptive binary quantization for image search with noisy labels

View at publisher Summarise with VIRAT
Authors: You Li, Ruozhen Ma
Cited by: 0
References: 12

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Supervised deep hashing models are often trained on large-scale datasets with manually assigned labels. However, obtaining such clean annotations is difficult in many real-world settings. In practice, datasets may contain mislabeled samples, which can noticeably impair retrieval performance. To address this issue, we propose a self-adaptive binary quantization (SABQ) framework that learns compact binary representations while remaining resistant to noisy labels. Specifically, we build a general deep hashing architecture and identify an early learning behavior that emerges under label noise. Based on this observation, we introduce a self-adaptive regularization term that strengthens the alignment between the model predictions and a temporally updated target. Our theoretical analysis shows that this regularization effectively reduces the adverse influence of noisy labels. Extensive experiments on both synthetic and real-world noisy datasets demonstrate that our method achieves superior performance compared with many existing approaches under various noise conditions.

Fields of Study & Subjects

Advanced Image and Video Retrieval Techniques Image Retrieval and Classification Techniques Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Algorithms, High Performance Computing and Artificial Intelligence (AHPCAI)
Pages
83–83
Publication date
19 March 2026
Language
EN