Simultaneous feature learning and hash coding with deep neural networks

By Hanjiang Lai, Yan Pan, Ye Liu, Shuicheng Yan. Computer Vision and Pattern Recognition, 2015

854 citations recorded by the source corpus.

Similarity-preserving hashing is a widely-used method for nearest neighbour search in large-scale image retrieval tasks. For most existing hashing methods, an image is first encoded as a vector of hand-engineering visual features, followed by another separate projection or quantization step that generates binary codes. However, such visual feature vectors may not be optimally compatible with the coding process, thus producing sub-optimal hashing codes. In this paper, we propose a deep architecture for supervised hashing, in which images are mapped into binary codes via carefully designed deep neural networks. The pipeline of the proposed deep architecture consists of three building blocks: 1) a sub-network with a stack of convolution layers to produce the effective intermediate image features; 2) a divide-and-encode module to divide the intermediate image features into multiple branches, each encoded into one hash bit; and 3) a triplet ranking loss designed to characterize that one image is more similar to the second image than to the third one. Extensive evaluations on several benchmark image datasets show that the proposed simultaneous feature learning and hash coding pipeline brings substantial improvements over other state-of-the-art supervised or unsupervised hashing methods.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/73543e2f04bc496930e5d5e9790559d4510e7596

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Fine-Grained Categorization and Dataset Bootstrapping Using Deep Metric Learning with Humans in the Loop (2015, background): “Recently, it has been successfully applied to variety of problems including face recognition and verification [39, 36], image retrieval [44], semantic hashing [29], product design [4], geo-localization [31] and style matching [41].”
Central Similarity Quantization for Efficient Image and Video Retrieval (2019, methodology): “DNNH [14] proposes to use a triplet ranking loss for similarity learning.”
Comparative Deep Learning of Hybrid Representations for Image Recommendations (2016, methodology): “Motivated by previous efforts on deep ranking models [14, 27, 30], we propose a comparative learning method to tackle the imperfect training data.”
Learning to Hash for Indexing Big Data—A Survey (2015, methodology): “The latest method called deep neural network hashing [109] goes beyond, in which the image representation and hash values are learned in one stage so that representation learning and hash learning are tightly coupled to benefit each other.”
Deep Cauchy Hashing for Hamming Space Retrieval (2018, methodology): “(2) Deep hashing methods DHN and HashNet learn less lossy hash codes by preserving the similarity information and controlling the quantization error, which also significantly outperform deep methods CNNH and DNNH without reducing the quantization error.”
Correlation Hashing Network for Efficient Cross-Modal Retrieval (2016, methodology): “Recent deep hashing methods [3, 5, 15, 32, 37] have given state of the art results, but they can only be used for single-modal retrieval.”
Learning in high-dimensional multimedia data: the state of the art (2015, methodology): “[47] proposed a “one-stage” supervised hashing method via a deep architecture that maps input images to binary codes.”
Feature Learning Based Deep Supervised Hashing with Pairwise Labels (2015, methodology): “Following [Lai et al., 2015], we only use the images associated with the 21 most frequent classes.”
Variational Deep Semantic Hashing for Text Documents (2017, background): “Most of the related work has focused on image data [4, 16, 19, 39] rather than text documents probably due to the effectiveness of the convolution neural networks (CNNs) to learn good low-dimensional representations of images.”
Towards Robust Deep Neural Networks with BANG (2016, background): “Machine learning is broadly used in various real-world vision applications and recent advances in deep learning have made deep neural networks the most powerful learning models that can be successfully applied to different vision problems [27, 25, 7, 28, 20, 14, 18, 17, 29].”
Pairwise Teacher-Student Network for Semi-Supervised Hashing (2019, background): “upervised hashing methods can capture more semantic information than unsupervised ones. Recently, with the rapid development of deep learning [11, 8], deep hashing methods have achieved great success [30, 12, 17, 34, 32, 14, 2]. It aims at learning hashcodes and the deep networks simultaneously, thu”
HashNet: Deep Learning to Hash by Continuation (2017, methodology): “Recently, deep learning to hash methods [37, 20, 31, 8, 40] have shown that end-to-end learning of feature representation and hash coding can be more effective using deep neural networks [18, 2], which can naturally encode any nonlinear hash functions.”
A Decade Survey of Content Based Image Retrieval Using Deep Learning (2020, methodology): “The triplet ranking loss is also utilized to incorporate the inter-class and intra-class differences in [63] for image retrieval.”
Cross-modal deep metric learning with multi-task regularization (2017, background): “On the one hand, triplet network has been proposed for single-modal metric learning [14, 15], which can model the relative similarity of images.”
Two-Stream Deep Hashing With Class-Specific Centers for Supervised Image Search (2020, background): “Network-in-network hashing (NINH) [53] proposes a deep architecture with three blocks: 1) a subnetwork to learn intermediate image features; 2) a divide-and-encode module that divides the intermediate image features into multiple branches, each encoded into one hash bit; and 3) a triplet loss to…”
Unsupervised Learning of Discriminative Attributes and Visual Representations (2016, methodology): “Our method outperforms the similar triplet hashing methods [27, 31], mainly because they lack the hash code constraints (in Eq.”