Abstract:Hashing has been drawing increasing attention in the task of large-scale image retrieval owing to its storage and computation efficiency, especially the recent asymmetric deep hashing methods. These approaches treat the query and database in an asymmetric way and can take full advantages of the whole training data. Though achieved the state-of-the-art performance, asymmetric deep hashing methods still suffer from the large quantization error and efficiency problem on large-scale dataset due to the tight coupling between query and database. In this paper, we propose a novel asymmetric hashing method, called D eep U ncoupled D iscrete H ashing (DUDH), for large-scale approximate nearest neighbor search. Instead of directly preserving the similarity between query and database, DUDH first exploits a small similarity-transfer image set to transfer the underlying semantic structures from database to query, and implicitly keep the desired similarity. As a result, the large similarity matrix is decomposed into two relatively small ones and query is decoupled from database. Then both database codes and similarity-transfer codes are directly learned during optimization. The quantization error of DUDH only exists in the process of preserving similarity between query and similarity-transfer set. By uncoupling query from database, the training cost of optimizing CNN model for query is no longer related to the size of database. Besides, to further accelerate the training process, we propose to optimize the similarity-transfer codes with a constant-approximation solution. In doing so, the training cost of optimizing similarity-transfer codes can be almost ignored. Extensive experiments on four widely used image retrieval benchmarks demonstrate that DUDH can achieve state-of-the-art retrieval performance with remarkable training cost reduction ( \(30\%-50\% \) relative).

Deep Hashing with Top Similarity Preserving for Image Retrieval

A Novel Deep Hashing Method with Top Similarity for Image Retrieval

Similarity and Ranking Preserving Deep Hashing for Image Retrieval.

Deep Top Similarity Hashing with Class-Wise Loss for Multi-Label Image Retrieval

Deep Hashing Network for Efficient Similarity Retrieval

Deep Hash Learning for Efficient Image Retrieval

Inductive Transfer Deep Hashing For Image Retrieval

Deep Semantic-Preserving and Ranking-Based Hashing for Image Retrieval.

Deep Self-Adaptive Hashing for Image Retrieval

Deep Semantic Reconstruction Hashing for Similarity Retrieval

Deep Self-Learning Hashing for Image Retrieval

Deep Balanced Discrete Hashing for Image Retrieval

Deep Supervised Hashing by Classification for Image Retrieval

Deep Fuzzy Hashing Network for Efficient Image Retrieval

Deep Uncoupled Discrete Hashing via Similarity Matrix Decomposition.

Deep Self-Taught Hashing for Image Retrieval

Ipdh: An Improved Pairwise-Based Deep Hashing Method For Large-Scale Image Retrieval

Deep Neighborhood Structure-Preserving Hashing for Large-Scale Image Retrieval

Unsupervised Deep Hashing with Fine-Grained Similarity-Preserving Contrastive Learning for Image Retrieval

Discrete Deep Hashing with Ranking Optimization for Image Retrieval

Deep Supervised Hashing with Hard Example Pairs Optimization for Image Retrieval