Supervised Hashing for Image Retrieval via Image Representation Learning
|
|
- Donna Lee
- 5 years ago
- Views:
Transcription
1 Supervised Hashing for Image Retrieval via Image Representation Learning Rongkai Xia, Yan Pan, Cong Liu (Sun Yat-Sen University) Hanjiang Lai, Shuicheng Yan (National University of Singapore)
2 Finding Similar Images Task: given a query image, find its nearest neighbors in an image database. search
3 Challenges Query image Image database features features features Challenge:how to efficiently search over millions or billions of data? e.g., if each image is represented by a 512-dim GIST vector, one needs 20G memory to store 10 million images. Nearest neighbors search
4 Query image Similarity-Preserving Hashing Image database features features feature Images are represented by binary codes. Efficient retrieval via bitwise operations. Space-efficient storage e.g., 10 million images in 32M memory, providing each image in 128-bit code Key question: how to preserve similarities?
5 Similarity-Preserving Hashing x 1 x 2 x 3
6 Similarity-Preserving Hashing x 1 dissimilar x 2 x 3 similar
7 Similarity-Preserving Hashing x 1 dissimilar x 2 x 3 Hashing h(x 1 ) 1 Hamming space (2-dim) h(x 2 ) Similarities are well preserved similar h(x 3 ) 0 1
8 Similarity-Preserving Hashing x 1 dissimilar x 2 x 3 Hashing h(x 1 ) 1 Hamming space (2-dim h(x 2 ) Similarities are well preserved similar h(x 3 ) Hashing 0 1 h(x 1 ) 1 Hamming space (2-dim h(x 3 ) 0 1 h(x 2 ) Similarities are not preserved
9 Related Work Long codes are needed to preserve similarities. Unsupervised Hashing LSH[Gionis et al. VLDB,1999] KLSH[Kulis and Grauman. PAMI,2012] SH[Weiss and Torralba. NIPS,2008] ITQ[Gong and Lazebnik. CVPR,2011] Learn compact codes by using label information. Semi-supervised Hashing Supervised Hashing SSH[Wang et al. CVPR,2010] MLH[Norouzi and Blei. ICML,2011] BRE[Kulis and Darrell. NIPS,2009] KSH[Liu et al. CVPR,2012] TSH[Lin et al. ICCV,2013]
10 Motivation In most existing methods, each image is firstly encoded by a vector of some handcrafted visual descriptor (e.g., GIST, BoW, SIFT) Concern: the chosen hand-crafted visual features do not necessarily guarantee to accurately preserve the semantic similarities of image pairs. e.g., a pair of semantically similar/dissimilar images may not have feature vectors with relatively small/large Euclidean distance.
11 Motivation In most existing methods, each image is firstly encoded by a vector of some handcrafted visual descriptor (e.g., GIST, BoW, SIFT). Concern: the chosen hand-crafted visual features do not necessarily guarantee to accurately preserve the semantic similarities of image pairs. e.g., a pair of semantically similar/dissimilar images may not have feature vectors with relatively small/large Euclidean distance. Feature extraction Gist vector Gist vector Gist vector Semantically similar images have large distance. Semantically dissimilar images have small distance.
12 Motivation In most existing methods, each image is firstly encoded by a vector of some handcrafted visual descriptor (e.g., GIST, BoW, SIFT) Concern: the chosen hand-crafted visual features do not necessarily guarantee to accurately preserve the semantic similarities of image pairs. e.g., a pair of semantically similar/dissimilar images may not have feature vectors with relatively small/large Euclidean distance. Hamming space (2-dim) Gist vector Feature extraction Gist vector Gist vector Hashing 1 Semantically similar images have large distance. Semantically dissimilar images have small distance. 0 1 Resulting in bad hash codes
13 Motivation A useful image representation is important in hash learning process. Hamming space (2-dim) Gist vector Feature extraction Gist vector Gist vector Hashing 1 Semantically similar images have large distance. Semantically dissimilar images have small distance. 0 1 Resulting in bad hash codes
14 Two-stage framework The Proposed Approach 1. Learn approximate hash codes for the training samples, i.e., the pairwise similarity matrix S is decomposed into a product where the ith row in H is the approximate hash code for the ith training image 2. By using the learnt H and the raw image pixels as input, learn image representation and hash functions via deep convolutional neural networks.
15 Two-stage framework The Proposed Approach Stage 1 S Similarity matrix H Approximate hash codes for the training samples
16 Two-stage framework The Proposed Approach S Similarity matrix H Approximate hash codes for the training samples Stage 2 Hash functions h h h 1 2 q Image representation
17 Stage 1: Learning Approximate Codes 3 training samples x 1 x 2 x 3 3-bit Hash codes x 1 x 2 x S H H T 1 min ( ) H n n T Sij HiH j i1 j1 q 1 T min S HH H q 2 F subject to : H { 1,1} nq 2 The Hamming distance of two hash codes has oneone correspondence to the inner product of these two codes. If images i and j are similar (dissimilar), the inner product of their approximate codes should be large (small). relaxation
18 Optimization Algorithm: random coordinate descent using Newton directions 1. Randomly select an entry H ij in H to update while keeping other entries fixed 2. Approximate the objective function by second-order Taylor expansion w.r.t. H ij Calculate a step-size d and update H ij by H H d ij ij 3. Repeat 1 and 2 until stopping criterion is satisfied
19 Optimization Algorithm: random coordinate descent using Newton directions 1. Randomly select an entry H ij in H to update while keeping other entries fixed H ij 2. Approximate the objective function by second-order Taylor expansion w.r.t. H ij Calculate a step-size d and update H ij by H H d ij ij 3. Repeat 1 and 2 until stopping criterion is satisfied
20 Optimization Algorithm: random coordinate descent using Newton directions 1. Randomly select an entry H ij in H to update while keeping other entries fixed 2. Approximate the objective function by second-order Taylor expansion w.r.t. H ij Calculate a step-size d and update H ij by H H d min g( H ) ( H H R ) H ij ij lj kj lk l1 k1 ( H R ) 2 ( H H R ) constant ij ii ij kj ik ki subject to : H [ 1,1] ij n n 1 g( H d) g( H ) g ( H ) d g ( H ) d 2 ' '' 2 ij ij ij ij 3. Repeat 1 and 2 until stopping criterion is satisfied 2 ij ij min gh ( d) d ij subject to : 1 H d 1 ij ( ) d max( 1 Hij,min(,1 Hij)) ( ) ' g Hij '' g Hij
21 Optimization Algorithm: random coordinate descent using Newton directions 1. Randomly select an entry H ij in H to update while keeping other entries fixed 2. Approximate the objective function by second-order Taylor expansion w.r.t. H ij Calculate a step-size d and update H ij by H H d ij ij 3. Repeat 1 and 2 until stopping criterion is satisfied The time complexity of the whole algorithm is O(Tqn^2) with small T, q. n: the number of training samples q: hash code length (less than 64 in our experiments) T: iterations (less than 5 in our experiments)
22 Stage 2: Learning Hash Functions S Similarity matrix H Approximate hash codes for the training samples Hash functions h h h 1 2 q + Image representation It is a multi-label binary classification problem that is solved by deep convolutional neural networks. It leans hash functions as well as image features. We propose two methods: CNNH and CNNH+.
23 Method 1: CNNH S Similarity matrix H Approximate hash codes for the training samples Hash functions h h h 1 2 q + Image representation Input Output Hash functions
24 Method 1: CNNH Input Prediction Output the hash code
25 Method2: CNNH+ beach, sky, (1,0,0,,1) Discrete class labels Use class labels to enhance performance. H Approximate hash codes for the training samples Hash functions h h h 1 2 q + Image representation
26 Method2: CNNH+ beach, sky, (1,0,0,,1) Discrete class labels Use class labels to enhance performance. H Approximate hash codes for the training samples Hash functions h h h 1 2 q + Image representation
27 Details of the Deep Convolutional Networks We adopt the architecture of [Krizhevsky, NIPS 2012] as our basic framework. Our network has three convolutional-pooling layers with rectified linear activation, max pooling and local contrast normalization, a standard fully connected layer, and an output layer with softmax activation. We use 32, 64, 128 filter (with the size 5*5) in the 1 st, 2 nd and 3 rd convolutional layer, respectively. We use dropout with a rate of 0.5.
28 Datasets MNIST: 70,000 greyscale images (in size 28*28) of handwritten digits from 0 to 9 CIFAR10: 60,000 color tinny images (in size 32*32) that are categorized in 10 classes NUS-WIDE: about 270,000 images collected from the web. It is a multi-label dataset.
29 Baseline Methods Unsupervised methods LSH [Gionis et al. VLDB,1999] SH [Weiss and Torralba. NIPS,2008] ITQ [Gong and Lazebnik. CVPR,2011] MLH [Norouzi and Blei. ICML,2011] BRE [Kulis and Darrell. NIPS,2009] ITQ-CCA [Gong and Lazebnik. CVPR,2011] KSH [Liu et al. CVPR,2012] Supervised methods
30 Evaluation Metrics Precision: Recall: precision recall Mean Average Precision (MAP): #{retrieved relevant images} #{retrieved images} #{retrieved relevant images} #{all relevant images} MAP 1 q i AP i AP P@ n I{image n is relevant} #{relevant images in top n result} n P@ n #{retrieved relevant image} n
31 Experimental Results MAP of Hamming ranking on MNIST w.r.t. different number of bits Methods 12-bit 24-bit 32-bit 48-bit CNNH CNNH KSH ITQ-CCA MLH BRE SH ITQ LSH relative increase of 8.2%~11.1%
32 Experimental Results MAP of Hamming ranking on CIFAR10 w.r.t. different number of bits Methods 算法 12-bit 位 24-bit 位 32-bit 位 48-bit 位 CNNH CNNH KSH ITQ-CCA MLH BRE SH ITQ LSH relative increase of 49.4%~54.6%
33 Experimental Results MAP of Hamming ranking on NUSWIDE w.r.t. different number of bits Methods 算法 12-bit 位 24-bit 位 32-bit 位 48-bit 位 CNNH CNNH KSH ITQ-CCA MLH BRE SH ITQ LSH relative increase of 6.3%~12.1%
34 Experimental Results Results on MNIST (a) precision within curves Hamming radius 2 (b) MAP curves within Hamming radius 2 (c) precision-recall curves with 48 bits (d) precision curves with 48 bits
35 Experimental Results Results on CIFAR-10 (a) precision curves within Hamming radius 2 (b) MAP curves within Hamming radius 2 (c) precision-recall curves with 48 bits (d) precision curves with 48 bits
36 Experimental Results Results on NUS-WIDE (a) precision curves within Hamming radius 2 (b) MAP curves within Hamming radius 2 (c) precision-recall curves with 48 bits (d) precision curves with 48 bits
37 Experimental Results CNNH+ vs. KSH with different hand-crafted features (a) Results on CIFAR-10 (b) Results on MNIST The performances of KSH with different features are inferior to those of CNNH+.
38 Thank you!
Supervised Hashing for Image Retrieval via Image Representation Learning
Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence Supervised Hashing for Image Retrieval via Image Representation Learning Rongkai Xia 1, Yan Pan 1, Hanjiang Lai 1,2, Cong Liu
More informationlearning stage (Stage 1), CNNH learns approximate hash codes for training images by optimizing the following loss function:
1 Query-adaptive Image Retrieval by Deep Weighted Hashing Jian Zhang and Yuxin Peng arxiv:1612.2541v2 [cs.cv] 9 May 217 Abstract Hashing methods have attracted much attention for large scale image retrieval.
More informationarxiv: v2 [cs.cv] 27 Nov 2017
Deep Supervised Discrete Hashing arxiv:1705.10999v2 [cs.cv] 27 Nov 2017 Qi Li Zhenan Sun Ran He Tieniu Tan Center for Research on Intelligent Perception and Computing National Laboratory of Pattern Recognition
More informationRongrong Ji (Columbia), Yu Gang Jiang (Fudan), June, 2012
Supervised Hashing with Kernels Wei Liu (Columbia Columbia), Jun Wang (IBM IBM), Rongrong Ji (Columbia), Yu Gang Jiang (Fudan), and Shih Fu Chang (Columbia Columbia) June, 2012 Outline Motivations Problem
More informationDeep Supervised Hashing with Triplet Labels
Deep Supervised Hashing with Triplet Labels Xiaofang Wang, Yi Shi, Kris M. Kitani Carnegie Mellon University, Pittsburgh, PA 15213 USA Abstract. Hashing is one of the most popular and powerful approximate
More informationDeep Semantic-Preserving and Ranking-Based Hashing for Image Retrieval
Proceedings of the Twenty-Fifth International Joint Conference on Artificial Intelligence (IJCAI-16) Deep Semantic-Preserving and Ranking-Based Hashing for Image Retrieval Ting Yao, Fuchen Long, Tao Mei,
More informationRECENT years have witnessed the rapid growth of image. SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval
SSDH: Semi-supervised Deep Hashing for Large Scale Image Retrieval Jian Zhang, Yuxin Peng, and Junchao Zhang arxiv:607.08477v [cs.cv] 28 Jul 206 Abstract The hashing methods have been widely used for efficient
More informationarxiv: v1 [cs.cv] 19 Oct 2017 Abstract
Improved Search in Hamming Space using Deep Multi-Index Hashing Hanjiang Lai and Yan Pan School of Data and Computer Science, Sun Yan-Sen University, China arxiv:70.06993v [cs.cv] 9 Oct 207 Abstract Similarity-preserving
More informationSupervised Hashing for Multi-labeled Data with Order-Preserving Feature
Supervised Hashing for Multi-labeled Data with Order-Preserving Feature Dan Wang, Heyan Huang (B), Hua-Kang Lin, and Xian-Ling Mao Beijing Institute of Technology, Beijing, China {wangdan12856,hhy63,1120141916,maoxl}@bit.edu.cn
More informationDeep Supervised Hashing for Fast Image Retrieval
Deep Supervised Hashing for Fast Image Retrieval Haomiao Liu 1,, Ruiping Wang 1, Shiguang Shan 1, Xilin Chen 1 1 Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences (CAS),
More informationLearning independent, diverse binary hash functions: pruning and locality
Learning independent, diverse binary hash functions: pruning and locality Ramin Raziperchikolaei and Miguel Á. Carreira-Perpiñán Electrical Engineering and Computer Science University of California, Merced
More informationHashing with Graphs. Sanjiv Kumar (Google), and Shih Fu Chang (Columbia) June, 2011
Hashing with Graphs Wei Liu (Columbia Columbia), Jun Wang (IBM IBM), Sanjiv Kumar (Google), and Shih Fu Chang (Columbia) June, 2011 Overview Graph Hashing Outline Anchor Graph Hashing Experiments Conclusions
More informationCompact Hash Code Learning with Binary Deep Neural Network
Compact Hash Code Learning with Binary Deep Neural Network Thanh-Toan Do, Dang-Khoa Le Tan, Tuan Hoang, Ngai-Man Cheung arxiv:7.0956v [cs.cv] 6 Feb 08 Abstract In this work, we firstly propose deep network
More informationBit-Scalable Deep Hashing with Regularized Similarity Learning for Image Retrieval and Person Re-identification
IEEE TRANSACTIONS ON IMAGE PROCESSING 1 Bit-Scalable Deep Hashing with Regularized Similarity Learning for Image Retrieval and Person Re-identification Ruimao Zhang, Liang Lin, Rui Zhang, Wangmeng Zuo,
More informationHashing with Binary Autoencoders
Hashing with Binary Autoencoders Ramin Raziperchikolaei Electrical Engineering and Computer Science University of California, Merced http://eecs.ucmerced.edu Joint work with Miguel Á. Carreira-Perpiñán
More informationApproximate Nearest Neighbor Search. Deng Cai Zhejiang University
Approximate Nearest Neighbor Search Deng Cai Zhejiang University The Era of Big Data How to Find Things Quickly? Web 1.0 Text Search Sparse feature Inverted Index How to Find Things Quickly? Web 2.0, 3.0
More informationIEEE TRANSACTIONS ON IMAGE PROCESSING 1. Instance-Aware Hashing for Multi-Label Image Retrieval
IEEE TRANSACTIONS ON IMAGE PROCESSING Instance-Aware Hashing for Multi-Label Image Retrieval Hanjiang Lai, Pan Yan, Xiangbo Shu,Yunchao Wei, Shuicheng Yan, Senior Member, IEEE arxiv:63.3234v [cs.cv] Mar
More informationThe Normalized Distance Preserving Binary Codes and Distance Table *
JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 32, XXXX-XXXX (2016) The Normalized Distance Preserving Binary Codes and Distance Table * HONGWEI ZHAO 1,2, ZHEN WANG 1, PINGPING LIU 1,2 AND BIN WU 1 1.
More informationover Multi Label Images
IBM Research Compact Hashing for Mixed Image Keyword Query over Multi Label Images Xianglong Liu 1, Yadong Mu 2, Bo Lang 1 and Shih Fu Chang 2 1 Beihang University, Beijing, China 2 Columbia University,
More informationAdaptive Binary Quantization for Fast Nearest Neighbor Search
IBM Research Adaptive Binary Quantization for Fast Nearest Neighbor Search Zhujin Li 1, Xianglong Liu 1*, Junjie Wu 1, and Hao Su 2 1 Beihang University, Beijing, China 2 Stanford University, Stanford,
More informationSupervised Hashing with Latent Factor Models
Supervised Hashing with Latent Factor Models Peichao Zhang Shanghai Key Laboratory of Scalable Computing and Systems Department of Computer Science and Engineering Shanghai Jiao Tong University, China
More informationColumn Sampling based Discrete Supervised Hashing
Column Sampling based Discrete Supervised Hashing Wang-Cheng Kang, Wu-Jun Li and Zhi-Hua Zhou National Key Laboratory for Novel Software Technology Collaborative Innovation Center of Novel Software Technology
More informationProgressive Generative Hashing for Image Retrieval
Progressive Generative Hashing for Image Retrieval Yuqing Ma, Yue He, Fan Ding, Sheng Hu, Jun Li, Xianglong Liu 2018.7.16 01 BACKGROUND the NNS problem in big data 02 RELATED WORK Generative adversarial
More informationarxiv: v1 [cs.cv] 30 Oct 2016
Accurate Deep Representation Quantization with Gradient Snapping Layer for Similarity Search Shicong Liu, Hongtao Lu arxiv:1610.09645v1 [cs.cv] 30 Oct 2016 Abstract Recent advance of large scale similarity
More informationRank Preserving Hashing for Rapid Image Search
205 Data Compression Conference Rank Preserving Hashing for Rapid Image Search Dongjin Song Wei Liu David A. Meyer Dacheng Tao Rongrong Ji Department of ECE, UC San Diego, La Jolla, CA, USA dosong@ucsd.edu
More informationHASHING has been recognized as an effective technique
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 1 Unsupervised Deep Hashing with Similarity-Adaptive and Discrete Optimization Fumin Shen, Yan Xu, Li Liu, Yang Yang, Zi Huang, Heng Tao Shen
More informationarxiv: v2 [cs.lg] 21 Apr 2016
Feature Learning based Deep Supervised Hashing with Pairwise Labels Wu-Jun Li, Sheng Wang and Wang-Cheng Kang National Key Laboratory for Novel Software Technology Department of Computer Science and Technology,
More informationarxiv: v3 [cs.cv] 10 Aug 2017
HOW SHOULD WE EVALUATE SUPERVISED HASHING? Alexandre Sablayrolles Matthijs Douze Nicolas Usunier Hervé Jégou Facebook AI Research arxiv:1609.06753v3 [cs.cv] 10 Aug 2017 ABSTRACT Hashing produces compact
More informationAsymmetric Deep Supervised Hashing
Asymmetric Deep Supervised Hashing Qing-Yuan Jiang and Wu-Jun Li National Key Laboratory for Novel Software Technology Collaborative Innovation Center of Novel Software Technology and Industrialization
More informationarxiv: v3 [cs.cv] 3 Dec 2017
Structured Deep Hashing with Convolutional Neural Networks for Fast Person Re-identification Lin Wu, Yang Wang ISSR, ITEE, The University of Queensland, Brisbane, QLD, 4072, Australia The University of
More informationHASHING has been a hot research topic on image processing. Bilinear Supervised Hashing Based on 2D Image Features
1 Bilinear Supervised Hashing Based on 2D Image Features Yujuan Ding, Wai Kueng Wong, Zhihui Lai and Zheng Zhang arxiv:1901.01474v1 [cs.cv] 5 Jan 2019 Abstract Hashing has been recognized as an efficient
More informationDeep Multi-Label Hashing for Large-Scale Visual Search Based on Semantic Graph
Deep Multi-Label Hashing for Large-Scale Visual Search Based on Semantic Graph Chunlin Zhong 1, Yi Yu 2, Suhua Tang 3, Shin ichi Satoh 4, and Kai Xing 5 University of Science and Technology of China 1,5,
More informationLocality-Constrained Deep Supervised Hashing for Image Retrieval
Locality-Constrained Deep Supervised Hashing for Image Retrieval Hao Zhu 1, Shenghua Gao 2 3M Cogent Beijing 1 School of Information Science and Technology, ShanghaiTech University 2 allenhaozhu@gmail.com,
More informationarxiv: v1 [cs.cv] 12 Aug 2016 Abstract
Deep Hashing: A Joint Approach for Image Signature Learning Yadong Mu Institute of Computer Science and Technology Peking University, China myd@pkueducn Zhu Liu Multimedia Department AT&T Labs, USA zliu@researhattcom
More informationDeep Cauchy Hashing for Hamming Space Retrieval
Deep Cauchy ashing for amming Space Retrieval Yue Cao, Mingsheng Long, Bin Liu, Jianmin Wang KLiss, MOE; School of Software, Tsinghua University, China National Engineering Laboratory for Big Data Software
More informationSupervised Hashing with Kernels
Supervised Hashing with Kernels Wei Liu Jun Wang Rongrong Ji Yu-Gang Jiang Shih-Fu Chang Electrical Engineering Department, Columbia University, New York, NY, USA IBM T. J. Watson Research Center, Yorktown
More informationLearning to Hash with Binary Reconstructive Embeddings
Learning to Hash with Binary Reconstructive Embeddings Brian Kulis and Trevor Darrell UC Berkeley EECS and ICSI Berkeley, CA {kulis,trevor}@eecs.berkeley.edu Abstract Fast retrieval methods are increasingly
More informationLarge-scale visual recognition Efficient matching
Large-scale visual recognition Efficient matching Florent Perronnin, XRCE Hervé Jégou, INRIA CVPR tutorial June 16, 2012 Outline!! Preliminary!! Locality Sensitive Hashing: the two modes!! Hashing!! Embedding!!
More informationSmart Hashing Update for Fast Response
Smart Hashing Update for Fast Response Qiang Yang, Longkai Huang, Wei-Shi Zheng*, and Yingbiao Ling School of Information Science and Technology, Sun Yat-sen University, China Guangdong Province Key Laboratory
More informationIterative Quantization: A Procrustean Approach to Learning Binary Codes
Iterative Quantization: A Procrustean Approach to Learning Binary Codes Yunchao Gong and Svetlana Lazebnik Department of Computer Science, UNC Chapel Hill, NC, 27599. {yunchao,lazebnik}@cs.unc.edu Abstract
More informationInstances on a Budget
Retrieving Similar or Informative Instances on a Budget Kristen Grauman Dept. of Computer Science University of Texas at Austin Work with Sudheendra Vijayanarasimham Work with Sudheendra Vijayanarasimham,
More informationIsometric Mapping Hashing
Isometric Mapping Hashing Yanzhen Liu, Xiao Bai, Haichuan Yang, Zhou Jun, and Zhihong Zhang Springer-Verlag, Computer Science Editorial, Tiergartenstr. 7, 692 Heidelberg, Germany {alfred.hofmann,ursula.barth,ingrid.haas,frank.holzwarth,
More informationLearning to Hash on Structured Data
Learning to Hash on Structured Data Qifan Wang, Luo Si and Bin Shen Computer Science Department, Purdue University West Lafayette, IN 47907, US wang868@purdue.edu, lsi@purdue.edu, bshen@purdue.edu Abstract
More informationDeep Hashing: A Joint Approach for Image Signature Learning
Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17) Deep Hashing: A Joint Approach for Image Signature Learning Yadong Mu, 1 Zhu Liu 2 1 Institute of Computer Science and
More informationon learned visual embedding patrick pérez Allegro Workshop Inria Rhônes-Alpes 22 July 2015
on learned visual embedding patrick pérez Allegro Workshop Inria Rhônes-Alpes 22 July 2015 Vector visual representation Fixed-size image representation High-dim (100 100,000) Generic, unsupervised: BoW,
More informationLearning Affine Robust Binary Codes Based on Locality Preserving Hash
Learning Affine Robust Binary Codes Based on Locality Preserving Hash Wei Zhang 1,2, Ke Gao 1, Dongming Zhang 1, and Jintao Li 1 1 Advanced Computing Research Laboratory, Beijing Key Laboratory of Mobile
More informationBilinear discriminant analysis hashing: A supervised hashing approach for high-dimensional data
Bilinear discriminant analysis hashing: A supervised hashing approach for high-dimensional data Author Liu, Yanzhen, Bai, Xiao, Yan, Cheng, Zhou, Jun Published 2017 Journal Title Lecture Notes in Computer
More informationLearning to Hash with Binary Reconstructive Embeddings
Learning to Hash with Binary Reconstructive Embeddings Brian Kulis Trevor Darrell Electrical Engineering and Computer Sciences University of California at Berkeley Technical Report No. UCB/EECS-2009-0
More informationSimilarity-Preserving Binary Hashing for Image Retrieval in large databases
Universidad Politécnica de Valencia Master s Final Project Similarity-Preserving Binary Hashing for Image Retrieval in large databases Author: Guillermo García Franco Supervisor: Dr. Roberto Paredes Palacios
More informationDiscrete Image Hashing Using Large Weakly Annotated Photo Collections
Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence (AAAI-16) Discrete Image Hashing Using Large Weakly Annotated Photo Collections Hanwang Zhang, Na Zhao, Xindi Shang, Huanbo Luan,
More informationHamming Distance Metric Learning
Hamming Distance Metric Learning Mohammad Norouzi David J. Fleet Ruslan Salakhutdinov, Departments of Computer Science and Statistics University of Toronto [norouzi,fleet,rsalakhu]@cs.toronto.edu Abstract
More informationDUe to the explosive increasing of data in real applications, Deep Discrete Supervised Hashing. arxiv: v1 [cs.
Deep Discrete Supervised Hashing Qing-Yuan Jiang, Xue Cui and Wu-Jun Li, Member, IEEE arxiv:707.09905v [cs.ir] 3 Jul 207 Abstract Hashing has been widely used for large-scale search due to its low storage
More informationAngular Quantization-based Binary Codes for Fast Similarity Search
Angular Quantization-based Binary Codes for Fast Similarity Search Yunchao Gong, Sanjiv Kumar, Vishal Verma, Svetlana Lazebnik Google Research, New York, NY, USA Computer Science Department, University
More informationBinary Embedding with Additive Homogeneous Kernels
Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-7) Binary Embedding with Additive Homogeneous Kernels Saehoon Kim, Seungjin Choi Department of Computer Science and Engineering
More informationHashGAN: Deep Learning to Hash with Pair Conditional Wasserstein GAN
HashGAN: Deep Learning to Hash with Pair Conditional Wasserstein GAN Yue Cao, Bin Liu, Mingsheng Long, Jianmin Wang KLiss, MOE; School of Software, Tsinghua University, China National Engineering Laboratory
More informationAsymmetric Discrete Graph Hashing
Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17) Asymmetric Discrete Graph Hashing Xiaoshuang Shi, Fuyong Xing, Kaidi Xu, Manish Sapkota, Lin Yang University of Florida,
More informationMachine Learning. Nonparametric methods for Classification. Eric Xing , Fall Lecture 2, September 12, 2016
Machine Learning 10-701, Fall 2016 Nonparametric methods for Classification Eric Xing Lecture 2, September 12, 2016 Reading: 1 Classification Representing data: Hypothesis (classifier) 2 Clustering 3 Supervised
More informationFace Video Retrieval via Deep Learning of Binary Hash Representations
Face Video Retrieval via Deep Learning of Binary Hash Representations Zhen Dong and Su Jia 2 and Tianfu Wu 3,4 and Mingtao Pei. Beijing Laboratory of IIT, School of Computer Science, Beijing Institute
More informationAdaptive Hash Retrieval with Kernel Based Similarity
Adaptive Hash Retrieval with Kernel Based Similarity Xiao Bai a, Cheng Yan a, Haichuan Yang a, Lu Bai b, Jun Zhou c, Edwin Robert Hancock d a School of Computer Science and Engineering, Beihang University,
More informationLocal Feature Binary Coding for Approximate Nearest Neighbor Search
LIU et al.: LOCAL FEATURE BINARY CODING 1 Local Feature Binary Coding for Approximate Nearest Neighbor Search Li Liu li2.liu@northumbria.ac.uk Mengyang Yu m.y.yu@ieee.org Ling Shao ling.shao@ieee.org Computer
More informationLearning Hash Codes with Listwise Supervision
2013 IEEE International Conference on Computer Vision Learning Hash Codes with Listwise Supervision Jun Wang Business Analytics and Mathematical Sciences IBM T. J. Watson Research Center wangjun@us.ibm.com
More informationProgressive Generative Hashing for Image Retrieval
Progressive Generative Hashing for Image Retrieval Yuqing Ma, Yue He, Fan Ding, Sheng Hu, Jun Li, Xianglong Liu State Key Lab of Software Development Environment, Beihang University, China mayuqing@nlsde.buaa.edu.cn,
More informationLarge scale object/scene recognition
Large scale object/scene recognition Image dataset: > 1 million images query Image search system ranked image list Each image described by approximately 2000 descriptors 2 10 9 descriptors to index! Database
More informationChapter 7. Learning Hypersurfaces and Quantisation Thresholds
Chapter 7 Learning Hypersurfaces and Quantisation Thresholds The research presented in this Chapter has been previously published in Moran (2016). 7.1 Introduction In Chapter 1 I motivated this thesis
More informationLearning Binary Codes and Binary Weights for Efficient Classification
Learning Binary Codes and Binary Weights for Efficient Classification Fu Shen, Yadong Mu, Wei Liu, Yang Yang, Heng Tao Shen University of Electronic Science and Technology of China AT&T Labs Research Didi
More informationProduct Quantization Network for Fast Image Retrieval
Product Quantization Network for Fast Image Retrieval Tan Yu 1, Junsong Yuan 2, Chen Fang 3, and Hailin Jin 3 1 Nanyang Technological University, Singapore tyu008@e.ntu.edu.sg 2 State University of New
More informationCluster Analysis. Mu-Chun Su. Department of Computer Science and Information Engineering National Central University 2003/3/11 1
Cluster Analysis Mu-Chun Su Department of Computer Science and Information Engineering National Central University 2003/3/11 1 Introduction Cluster analysis is the formal study of algorithms and methods
More informationEvaluation of Hashing Methods Performance on Binary Feature Descriptors
Evaluation of Hashing Methods Performance on Binary Feature Descriptors Jacek Komorowski and Tomasz Trzcinski 2 Warsaw University of Technology, Warsaw, Poland jacek.komorowski@gmail.com 2 Warsaw University
More information2022 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 19, NO. 9, SEPTEMBER 2017
2022 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 19, NO. 9, SEPTEMBER 2017 Asymmetric Binary Coding for Image Search Fu Shen, Yang Yang, Li Liu, Wei Liu, Dacheng Tao, Fellow, IEEE, and Heng Tao Shen Abstract
More informationClassification-Based Supervised Hashing with Complementary Networks for Image Search
STUDENT, PROF, COLLABORATOR: BMVC AUTHOR GUIDELINES 1 Classification-Based Supervised Hashing with Complementary Networks for Image Search Dong-ju Jeong jeongdj@ispl.snu.ac.kr Sungkwon Choo chewry@ispl.snu.ac.kr
More informationLearning Discrete Representations via Information Maximizing Self-Augmented Training
A. Relation to Denoising and Contractive Auto-encoders Our method is related to denoising auto-encoders (Vincent et al., 2008). Auto-encoders maximize a lower bound of mutual information (Cover & Thomas,
More informationarxiv: v1 [cs.lg] 22 Feb 2016
Noname manuscript No. (will be inserted by the editor) Structured Learning of Binary Codes with Column Generation Guosheng Lin Fayao Liu Chunhua Shen Jianxin Wu Heng Tao Shen arxiv:1602.06654v1 [cs.lg]
More informationReciprocal Hash Tables for Nearest Neighbor Search
Reciprocal Hash Tables for Nearest Neighbor Search Xianglong Liu and Junfeng He and Bo Lang State Key Lab of Software Development Environment, Beihang University, Beijing, China Department of Electrical
More informationDeep Neural Networks:
Deep Neural Networks: Part II Convolutional Neural Network (CNN) Yuan-Kai Wang, 2016 Web site of this course: http://pattern-recognition.weebly.com source: CNN for ImageClassification, by S. Lazebnik,
More informationDeep Learning for Computer Vision II
IIIT Hyderabad Deep Learning for Computer Vision II C. V. Jawahar Paradigm Shift Feature Extraction (SIFT, HoG, ) Part Models / Encoding Classifier Sparrow Feature Learning Classifier Sparrow L 1 L 2 L
More informationCLSH: Cluster-based Locality-Sensitive Hashing
CLSH: Cluster-based Locality-Sensitive Hashing Xiangyang Xu Tongwei Ren Gangshan Wu Multimedia Computing Group, State Key Laboratory for Novel Software Technology, Nanjing University xiangyang.xu@smail.nju.edu.cn
More informationSemi-Supervised Deep Hashing with a Bipartite Graph
Semi-Supervised Deep Hashing with a Bipartite Graph Xinyu Yan, Lijun Zhang, Wu-Jun Li National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China {yanxy, zhanglj}@lamda.nju.edu.cn,
More informationScalable Graph Hashing with Feature Transformation
Proceedings of the Twenty-ourth International Joint Conference on Artificial Intelligence (IJCAI 2015) Scalable Graph Hashing with eature Transformation Qing-Yuan Jiang and Wu-Jun Li National Key Laboratory
More informationarxiv: v2 [cs.cv] 27 Oct 2018
SUPERVISED HASHING WITH END-TO-END BINARY DEEP NEURAL NETWORK Dang-Khoa Le Tan Thanh-Toan Do Ngai-Man Cheung Singapore University of Technology and Design (SUTD) The University of Adelaide arxiv:1711.08901v
More informationComputer Vision Lecture 16
Computer Vision Lecture 16 Deep Learning for Object Categorization 14.01.2016 Bastian Leibe RWTH Aachen http://www.vision.rwth-aachen.de leibe@vision.rwth-aachen.de Announcements Seminar registration period
More informationFast Supervised Hashing with Decision Trees for High-Dimensional Data
Fast Supervised Hashing with Decision Trees for High-Dimensional Data Guosheng Lin, Chunhua Shen, Qinfeng Shi, Anton van den Hengel, David Suter The University of Adelaide, SA 5005, Australia Abstract
More informationMulti-View Visual Recognition of Imperfect Testing Data
Multi-View Visual Recognition of Imperfect Testing Data MM'15, October 26 30, 2015, Brisbane, Australia Qilin Zhang Stevens Institute of Technology 1 Castle Point Terrace Hoboken, NJ, USA, 07030 Gang Hua
More informationBoosting Complementary Hash Tables for Fast Nearest Neighbor Search
Boosting Complementary Hash Tables for Fast Nearest Neighbor Search Xianglong Liu Cheng Deng Yadong Mu Zhujin Li State Key Lab of Software Development Environment, Beihang University, Being 1191, China
More informationProceedings of the International MultiConference of Engineers and Computer Scientists 2018 Vol I IMECS 2018, March 14-16, 2018, Hong Kong
, March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong , March 14-16, 2018, Hong Kong TABLE I CLASSIFICATION ACCURACY OF DIFFERENT PRE-TRAINED MODELS ON THE TEST DATA
More informationarxiv: v1 [cs.cv] 19 Jul 2016
arxiv:1607.05529v1 [cs.cv] 19 Jul 2016 Dual Purpose Hashing Haomiao Liu 1,2, Ruiping Wang 1, Shiguang Shan 1, Xilin Chen 1 1 Key Lab of Intelligent Information Processing of Chinese Academy of Sciences(CAS),
More informationComplementary Projection Hashing
23 IEEE International Conference on Computer Vision Complementary Projection Hashing Zhongg Jin, Yao Hu, Yue Lin, Debing Zhang, Shiding Lin 2, Deng Cai, Xuelong Li 3 State Key Lab of CAD&CG, College of
More informationSearching in one billion vectors: re-rank with source coding
Searching in one billion vectors: re-rank with source coding Hervé Jégou INRIA / IRISA Romain Tavenard Univ. Rennes / IRISA Laurent Amsaleg CNRS / IRISA Matthijs Douze INRIA / LJK ICASSP May 2011 LARGE
More information442 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 15, NO. 2, FEBRUARY 2013
442 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 15, NO. 2, FEBRUARY 2013 Query-Adaptive Image Search With Hash Codes Yu-Gang Jiang, Jun Wang, Member, IEEE, Xiangyang Xue, Member, IEEE, and Shih-Fu Chang, Fellow,
More informationLearning Visual Semantics: Models, Massive Computation, and Innovative Applications
Learning Visual Semantics: Models, Massive Computation, and Innovative Applications Part II: Visual Features and Representations Liangliang Cao, IBM Watson Research Center Evolvement of Visual Features
More informationDeep Asymmetric Pairwise Hashing
Deep Asymmetric Pairwise Hashing Fumin Shen, Xin Gao, Li Liu 2, Yang Yang, Heng Tao Shen Center for Future Media and School of Computer Science and Engineering, University of Electronic Science and Technology
More informationDeep Semantic Hashing with Generative Adversarial Networks *
Deep Semantic Hashing with Generative Adversarial Networks * Zhaofan Qiu University of Science and Technology of China Hefei, China zhaofanqiu@gmail.com ABSTRACT Ting Yao Microsoft Research Asia Beijing,
More informationarxiv: v1 [cs.cv] 20 Dec 2016
End-to-End Pedestrian Collision Warning System based on a Convolutional Neural Network with Semantic Segmentation arxiv:1612.06558v1 [cs.cv] 20 Dec 2016 Heechul Jung heechul@dgist.ac.kr Min-Kook Choi mkchoi@dgist.ac.kr
More informationarxiv: v1 [cs.cv] 23 Apr 2018
Deep Semantic Hashing with Generative Adversarial Networks arxiv:1804.08275v1 [cs.cv] 23 Apr 2018 ABSTRACT Zhaofan Qiu University of Science and Technology of China Hefei, China zhaofanqiu@gmail.com Ting
More informationFast Indexing and Search. Lida Huang, Ph.D. Senior Member of Consulting Staff Magma Design Automation
Fast Indexing and Search Lida Huang, Ph.D. Senior Member of Consulting Staff Magma Design Automation Motivation Object categorization? http://www.cs.utexas.edu/~grauman/slides/jain_et_al_cvpr2008.ppt Motivation
More informationA Sparse and Locally Shift Invariant Feature Extractor Applied to Document Images
A Sparse and Locally Shift Invariant Feature Extractor Applied to Document Images Marc Aurelio Ranzato Yann LeCun Courant Institute of Mathematical Sciences New York University - New York, NY 10003 Abstract
More informationComplementary Hashing for Approximate Nearest Neighbor Search
Complementary Hashing for Approximate Nearest Neighbor Search Hao Xu Jingdong Wang Zhu Li Gang Zeng Shipeng Li Nenghai Yu MOE-MS KeyLab of MCC, University of Science and Technology of China, P. R. China
More informationOptimizing affinity-based binary hashing using auxiliary coordinates
Optimizing affinity-based binary hashing using auxiliary coordinates Ramin Raziperchiolaei Miguel Á. Carreira-Perpiñán Electrical Engineering and Computer Science, University of California, Merced http://eecs.ucmerced.edu
More informationDeepIndex for Accurate and Efficient Image Retrieval
DeepIndex for Accurate and Efficient Image Retrieval Yu Liu, Yanming Guo, Song Wu, Michael S. Lew Media Lab, Leiden Institute of Advance Computer Science Outline Motivation Proposed Approach Results Conclusions
More informationLocality-Sensitive Codes from Shift-Invariant Kernels Maxim Raginsky (Duke) and Svetlana Lazebnik (UNC)
Locality-Sensitive Codes from Shift-Invariant Kernels Maxim Raginsky (Duke) and Svetlana Lazebnik (UNC) Goal We want to design a binary encoding of data such that similar data points (similarity measures
More informationSemi-Supervised Hashing for Scalable Image Retrieval
Semi-Supervised Hashing for Scalable Image Retrieval Jun Wang Columbia University New Yor, NY, 10027 jwang@ee.columbia.edu Sanjiv Kumar Google Research New Yor, NY, 10011 sanjiv@google.com Shih-Fu Chang
More information