DistillHash: Unsupervised Deep Hashing by Distilling Data Pairs

Yang, Erkun; Liu, Tongliang; Deng, Cheng; Liu, Wei; Tao, Dacheng

doi:10.1109/cvpr.2019.00306

Cited by 127 publications

(71 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then, SSDH constructed a similarity matrix and designed a pairwise loss function to preserve this semantic information. DistillHash [24] treated the initial similarity relationship as noisy labels and learned a distilled data set to perform unsupervised hash codes learning.…”

Section: Related Work a Deep Hashing Methodsmentioning

confidence: 99%

“…Similar to [22], [24], we aim to decrease the noisy relationship in unsupervised deep hashing for better retrieval accuracy. However, different from existing unsupervised methods which use the overall information of the images to estimate the similarity matrix, we propose multi-part corresponding FIGURE 2.…”

Section: Related Work a Deep Hashing Methodsmentioning

confidence: 99%

“…Unsupervised deep hashing methods [19]- [22], [24], [25], which is the focus of this paper, bring an opportunity to relieve the pressure of labeling. But how to transform high-dimensional images into low-dimensional hash codes without labels and effectively maintain the similarity between images has become a new issue in this field.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Pseudo Labels and Soft Multi-Part Corresponding Similarity for Unsupervised Deep Hashing

Xie

et al. 2020

IEEE Access

View full text Add to dashboard Cite

In recent years, unsupervised deep hashing methods have achieved great success in large-scale image retrieval. However, these approaches still suffer two major problems in real world applications. On the one hand, due to the lack of effective supervision information, hash codes of different categories are easily judged to be similar. On the other hand, binary semantic similarity matrices can not reflect ranking relationship and the internal structure information of different images. To solve these problems, we propose a novel unsupervised deep hashing method, named Pseudo labels and Soft multi-part Corresponding similarity based Hashing (PSCH), to ensure the heterogeneity of the hash codes. Specifically, we propose a ''pseudo labels'' method that use k-means clustering and a distance threshold to generate the pseudo labels. Further, in order to reflect the hash codes similarity between instances within the same class, we propose a novel soft multi-part corresponding similarity method to learn better hash codes. This method can divide deep feature maps into several groups and compute the attention map for multi-part similarity matrices. In addition, a novel loss function is proposed to support learning with pseudo labels and soft multi-part corresponding similarity for achieving better performance. Comprehensive experiments on CIFAR-10, NUSWIDE, and Flickr demonstrate that our method can generate high-quality hash codes and outperform state-of-the-art unsupervised hashing methods by a large margin.

show abstract

Section: Related Work a Deep Hashing Methodsmentioning

confidence: 99%

Section: Related Work a Deep Hashing Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Pseudo Labels and Soft Multi-Part Corresponding Similarity for Unsupervised Deep Hashing

Xie

et al. 2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…In [72], the proposed unsupervised hashing framework unifies the quantization error minimization, likelihood and mutual information maximization to preserve the feature distribution for better code quality. In DistillHash [66], Bayesian learning framework is integrated into the hash code learning, where a distilled data set is investigated automatically and further utilized to learn the compact binary code. In [51], the proposed DVB adopts a conditional auto-encoding variational Bayesian networks to estimate the training data structure under the probabilistic inference process with hashing objectives, thus improving the code quality.…”

Section: B Learning-based Feature Descriptorsmentioning

confidence: 99%

On Aggregation of Unsupervised Deep Binary Descriptor With Weak Bits

Lin

Ding

et al. 2020

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

Despite the thrilling success achieved by existing binary descriptors, most of them are still in the mire of three limitations: 1) vulnerable to the geometric transformations; 2) incapable of preserving the manifold structure when learning binary codes; 3) NO guarantee to find the true match if multiple candidates happen to have the same Hamming distance to a given query. All these together make the binary descriptor less effective when handling large-scale visual recognition tasks. In this paper, we propose a novel learning-based feature descriptor, namely Unsupervised Deep Binary Descriptor (UDBD), which learns transformation invariant binary descriptors via projecting the original data and their transformed sets into a joint binary space. Moreover, we involve a 2,1-norm loss term in the binary embedding process to gain simultaneously the robustness against data noises and less probability of mistakenly flipping bits of the binary descriptor, on top of it, a graph constraint is used to preserve the original manifold structure in the binary space. Furthermore, a weak bit mechanism is adopted to find the real match from candidates sharing the same minimum Hamming distance, thus enhancing matching performance. Extensive experimental results on public datasets show the superiority of UDBD in terms of matching and retrieval accuracy over the state-of-the-arts.

show abstract

“…Cross-modal similarity retrieval has been a popular research topic [14,18,20,25,28,28,32,33,37,40] with the objective to search the semantic similar instances from different modalities. In a typical scenario, instances in one modality, e.g., images, are retrieved given a query from another modality, e.g., text.…”

Section: Introductionmentioning

confidence: 99%

Joint-modal Distribution-based Similarity Hashing for Large-scale Unsupervised Deep Cross-modal Retrieval

Song

Qian

Guan

et al. 2020

Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval

119

View full text Add to dashboard Cite

Hashing-based cross-modal search which aims to map multiple modality features into binary codes has attracted increasingly attention due to its storage and search efficiency especially in largescale database retrieval. Recent unsupervised deep cross-modal hashing methods have shown promising results. However, existing approaches typically suffer from two limitations: (1) They usually learn cross-modal similarity information separately or in a redundant fusion manner, which may fail to capture semantic correlations among instances from different modalities sufficiently and effectively. (2) They seldom consider the sampling and weighting schemes for unsupervised cross-modal hashing, resulting in the lack of satisfactory discriminative ability in hash codes. To overcome these limitations, we propose a novel unsupervised deep cross-modal hashing method called Joint-modal Distributionbased Similarity Hashing (JDSH) for large-scale cross-modal retrieval. First, we propose a novel cross-modal joint-training method by constructing a joint-modal similarity matrix to fully preserve the cross-modal semantic correlations among instances. Second, we propose a sampling and weighting scheme termed the Distributionbased Similarity Decision and Weighting (DSDW) method for unsupervised cross-modal hashing, which is able to generate more discriminative hash codes by pushing semantic similar instance pairs closer and pulling semantic dissimilar instance pairs apart. The experimental results demonstrate the superiority of JDSH compared with several unsupervised cross-modal hashing methods on two public datasets NUS-WIDE and MIRFlickr.

show abstract

DistillHash: Unsupervised Deep Hashing by Distilling Data Pairs

Cited by 127 publications

References 55 publications

Pseudo Labels and Soft Multi-Part Corresponding Similarity for Unsupervised Deep Hashing

Pseudo Labels and Soft Multi-Part Corresponding Similarity for Unsupervised Deep Hashing

On Aggregation of Unsupervised Deep Binary Descriptor With Weak Bits

Joint-modal Distribution-based Similarity Hashing for Large-scale Unsupervised Deep Cross-modal Retrieval

Contact Info

Product

Resources

About