deploy: b78a999

index.html

@@ -154,7 +154,6 @@ <h3 id="-browse-papers-by-tag">🏷 Browse Papers by Tag</h3>
 <tag><a href="/tags.html#CVPR">CVPR</a></tag>
 <tag><a href="/tags.html#CVPRW">CVPRW</a></tag>
 <tag><a href="/tags.html#Case Study">Case Study</a></tag>
-<tag><a href="/tags.html#Cross Modal">Cross Modal</a></tag>
 <tag><a href="/tags.html#Cross-Modal">Cross-Modal</a></tag>
 <tag><a href="/tags.html#Dataset">Dataset</a></tag>
 <tag><a href="/tags.html#Deep Learning">Deep Learning</a></tag>

paper-abstracts.json

@@ -163,7 +163,7 @@
 {"key": "su2019unsupervised", "year": "2019", "title":"Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval", "abstract": "<p><img src=\"https://github.com/zzs1994/DJSRH/blob/master/page_image/DJRSH.png?raw=true\" alt=\"Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval\" title=\"Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval\" /></p>\n\n<p>Cross-modal hashing encodes the multimedia data into a common binary hash space in which the correlations among the samples from different modalities can be effectively measured. Deep cross-modal hashing further improves the retrieval performance as the deep neural networks can generate more semantic relevant features and hash codes. In this paper, we study the unsupervised deep cross-modal hash coding and propose Deep Joint Semantics Reconstructing Hashing (DJSRH), which has the following two main advantages. First, to learn binary codes that preserve the neighborhood structure of the original data, DJSRH constructs a novel joint-semantics affinity matrix which elaborately integrates the original neighborhood information from different modalities and accordingly is capable to capture the latent intrinsic semantic affinity for the input multi-modal instances. Second, DJSRH later trains the networks to generate binary codes that maximally reconstruct above joint-semantics relations via the proposed reconstructing framework, which is more competent for the batch-wise training as it reconstructs the specific similarity value unlike the common Laplacian constraint merely preserving the similarity order. Extensive experiments demonstrate the significant improvement by DJSRH in various cross-modal retrieval tasks.</p>\n", "tags": ["Cross-Modal","Unsupervised","ICCV","Has Code","Deep Learning"] },
 {"key": "subramanya2019diskann", "year": "2019", "title":"DiskANN: Fast Accurate Billion-point Nearest Neighbor Search on a Single Node", "abstract": "<p>Current state-of-the-art approximate nearest neighbor search (ANNS) algorithms generate indices that must be stored in main memory for fast high-recall search. This makes them expensive and limits the size of the dataset. We present a new graph-based indexing and search system called DiskANN that can index, store, and search a billion point database on a single workstation with just 64GB RAM and an inexpensive solid-state drive (SSD). Contrary to current wisdom, we demonstrate that the SSD-based indices built by DiskANN can meet all three desiderata for large-scale ANNS: high-recall, low query latency and high density (points indexed per node). On the billion point SIFT1B bigann dataset, DiskANN serves &gt; 5000 queries a second with &lt; 3ms mean latency and 95%+ 1-recall@1 on a 16 core machine, where state-of-the-art billion-point ANNS algorithms with similar memory footprint like FAISS and IVFOADC+G+P plateau at around 50% 1-recall@1. Alternately, in the high recall regime, DiskANN can index and serve 5 − 10x more points per node compared to state-of-the-art graph- based methods such as HNSW and NSG. Finally, as part of our overall DiskANN system, we introduce Vamana, a new graph-based ANNS index that is more versatile than the graph indices even for in-memory indices.</p>\n\n", "tags": ["NeurIPS","Graph","Unsupervised"] },
 {"key": "sun2019supervised", "year": "2019", "title":"Supervised Hierarchical Cross-Modal Hashing", "abstract": "<p>Recently, due to the unprecedented growth of multimedia data,\ncross-modal hashing has gained increasing attention for the\nefficient cross-media retrieval. Typically, existing methods on crossmodal hashing treat labels of one instance independently but\noverlook the correlations among labels. Indeed, in many real-world\nscenarios, like the online fashion domain, instances (items) are\nlabeled with a set of categories correlated by certain hierarchy. In\nthis paper, we propose a new end-to-end solution for supervised\ncross-modal hashing, named HiCHNet, which explicitly exploits the\nhierarchical labels of instances. In particular, by the pre-established\nlabel hierarchy, we comprehensively characterize each modality\nof the instance with a set of layer-wise hash representations. In\nessence, hash codes are encouraged to not only preserve the layerwise semantic similarities encoded by the label hierarchy, but also\nretain the hierarchical discriminative capabilities. Due to the lack\nof benchmark datasets, apart from adapting the existing dataset\nFashionVC from fashion domain, we create a dataset from the\nonline fashion platform Ssense consisting of 15, 696 image-text\npairs labeled by 32 hierarchical categories. Extensive experiments\non two real-world datasets demonstrate the superiority of our model\nover the state-of-the-art methods.</p>\n", "tags": ["SIGIR","Cross-Modal","Deep Learning"] },
-{"key": "sun2022deep", "year": "2022", "title":"Deep Normalized Cross-Modal Hashing With Bi-Direction Relation Reasoning", "abstract": "<p>Due to the continuous growth of large-scale multi-modal data and increasing requirements for retrieval speed, deep cross-modal hashing has gained increasing attention recently. Most of existing studies take a similarity matrix as supervision to optimize their models, and the inner product between continuous surrogates of hash codes is utilized to depict the similarity in the Hamming space. However, all of them merely consider the relevant information to build the similarity matrix, ignoring the contribution of the irrelevant one, i.e., the categories that samples do not belong to. Therefore, they cannot effectively alleviate the effect of dissimilar samples. Moreover, due to the modality distribution difference, directly utilizing continuous surrogates of hash codes to calculate similarity may induce suboptimal retrieval performance. To tackle these issues, in this paper, we propose a novel deep normalized cross-modal hashing scheme with bi-direction relation reasoning, named Bi_NCMH. Specifically, we build the multi-level semantic similarity matrix by considering bi-direction relation, i.e., consistent and inconsistent relation. It hence can holistically characterize relations among instances. Besides, we execute feature normalization on continuous surrogates of hash codes to eliminate the deviation caused by modality gap, which further reduces the negative impact of binarization on retrieval performance. Extensive experiments on two cross-modal benchmark datasets demonstrate the superiority of our model over several state-of-the-art baselines.</p>\n", "tags": ["CVPR","Cross Modal","Deep Learning"] },
+{"key": "sun2022deep", "year": "2022", "title":"Deep Normalized Cross-Modal Hashing With Bi-Direction Relation Reasoning", "abstract": "<p>Due to the continuous growth of large-scale multi-modal data and increasing requirements for retrieval speed, deep cross-modal hashing has gained increasing attention recently. Most of existing studies take a similarity matrix as supervision to optimize their models, and the inner product between continuous surrogates of hash codes is utilized to depict the similarity in the Hamming space. However, all of them merely consider the relevant information to build the similarity matrix, ignoring the contribution of the irrelevant one, i.e., the categories that samples do not belong to. Therefore, they cannot effectively alleviate the effect of dissimilar samples. Moreover, due to the modality distribution difference, directly utilizing continuous surrogates of hash codes to calculate similarity may induce suboptimal retrieval performance. To tackle these issues, in this paper, we propose a novel deep normalized cross-modal hashing scheme with bi-direction relation reasoning, named Bi_NCMH. Specifically, we build the multi-level semantic similarity matrix by considering bi-direction relation, i.e., consistent and inconsistent relation. It hence can holistically characterize relations among instances. Besides, we execute feature normalization on continuous surrogates of hash codes to eliminate the deviation caused by modality gap, which further reduces the negative impact of binarization on retrieval performance. Extensive experiments on two cross-modal benchmark datasets demonstrate the superiority of our model over several state-of-the-art baselines.</p>\n", "tags": ["CVPR","Cross-Modal","Deep Learning"] },
 {"key": "sundaram2013streaming", "year": "2013", "title":"Streaming Similarity Search over one Billion Tweets using Parallel Locality-Sensitive Hashing", "abstract": "<p>Finding nearest neighbors has become an important operation on databases, with applications to text search, multimedia indexing,\nand many other areas. One popular algorithm for similarity search, especially for high dimensional data (where spatial indexes like kdtrees do not perform well) is Locality Sensitive Hashing (LSH), an\napproximation algorithm for finding similar objects. In this paper, we describe a new variant of LSH, called Parallel\nLSH (PLSH) designed to be extremely efficient, capable of scaling out on multiple nodes and multiple cores, and which supports highthroughput streaming of new data. Our approach employs several\nnovel ideas, including: cache-conscious hash table layout, using a 2-level merge algorithm for hash table construction; an efficient\nalgorithm for duplicate elimination during hash-table querying; an insert-optimized hash table structure and efficient data expiration\nalgorithm for streaming data; and a performance model that accurately estimates performance of the algorithm and can be used to\noptimize parameter settings. We show that on a workload where we perform similarity search on a dataset of &gt; 1 Billion tweets, with\nhundreds of millions of new tweets per day, we can achieve query times of 1–2.5 ms. We show that this is an order of magnitude faster\nthan existing indexing schemes, such as inverted indexes. To the best of our knowledge, this is the fastest implementation of LSH,\nwith table construction times up to 3.7x faster and query times that are 8.3x faster than a basic implementation.</p>\n", "tags": ["VLDB","LSH","Streaming Data"] },
 {"key": "tiny2008million", "year": "2008", "title":"80 million tiny images: a large dataset for non-parametric object and scene recognition", "abstract": "<p>With the advent of the Internet, billions of images\nare now freely available online and constitute a dense sampling\nof the visual world. Using a variety of non-parametric methods,\nwe explore this world with the aid of a large dataset of 79,302,017\nimages collected from the Web. Motivated by psychophysical\nresults showing the remarkable tolerance of the human visual\nsystem to degradations in image resolution, the images in the\ndataset are stored as 32 × 32 color images. Each image is\nloosely labeled with one of the 75,062 non-abstract nouns in\nEnglish, as listed in the Wordnet lexical database. Hence the\nimage database gives a comprehensive coverage of all object\ncategories and scenes. The semantic information from Wordnet\ncan be used in conjunction with nearest-neighbor methods to\nperform object classification over a range of semantic levels\nminimizing the effects of labeling noise. For certain classes that\nare particularly prevalent in the dataset, such as people, we are\nable to demonstrate a recognition performance comparable to\nclass-specific Viola-Jones style detectors.</p>\n", "tags": [] },
 {"key": "wang2010semisupervised", "year": "2010", "title":"Semi-supervised hashing for scalable image retrieval", "abstract": "<p>Large scale image search has recently attracted considerable\nattention due to easy availability of huge amounts of\ndata. Several hashing methods have been proposed to allow\napproximate but highly efficient search. Unsupervised\nhashing methods show good performance with metric distances\nbut, in image search, semantic similarity is usually\ngiven in terms of labeled pairs of images. There exist supervised\nhashing methods that can handle such semantic similarity\nbut they are prone to overfitting when labeled data\nis small or noisy. Moreover, these methods are usually very\nslow to train. In this work, we propose a semi-supervised\nhashing method that is formulated as minimizing empirical\nerror on the labeled data while maximizing variance\nand independence of hash bits over the labeled and unlabeled\ndata. The proposed method can handle both metric as\nwell as semantic similarity. The experimental results on two\nlarge datasets (up to one million samples) demonstrate its\nsuperior performance over state-of-the-art supervised and\nunsupervised methods.</p>\n", "tags": ["CVPR","Supervised","Image Retrieval"] },

publications-metadata/andoni2006near.json

@@ -1 +1 @@
-[["yandextexttoimage1B", "Yandex Text-to-Image-1B"], ["jin2021unsupervised", "Unsupervised Discrete Hashing with Affinity Similarity"], ["yuan2020quant", "Central Similarity Quantization for Efficient Image and Video Retrieval"], ["sun2022deep", "Deep Normalized Cross-Modal Hashing With Bi-Direction Relation Reasoning"]]
+[["yandextexttoimage1B", "Yandex Text-to-Image-1B"], ["jin2021unsupervised", "Unsupervised Discrete Hashing with Affinity Similarity"], ["sun2022deep", "Deep Normalized Cross-Modal Hashing With Bi-Direction Relation Reasoning"], ["yuan2020quant", "Central Similarity Quantization for Efficient Image and Video Retrieval"]]

publications-metadata/sun2022deep.json

@@ -1 +1 @@
-[["su2019unsupervised", "Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval"], ["wang2019semi", "Semi-supervised Deep Quantization for Cross-modal Search"], ["liu2020joint", "Joint-modal Distribution-based Similarity Hashing for Large-scale Unsupervised Deep Cross-modal Retrieval"], ["chen2024supervised", "Supervised Consensus Anchor Graph Hashing for Cross Modal Retrieval"]]
+[["su2019unsupervised", "Deep Joint-Semantics Reconstructing Hashing for Large-Scale Unsupervised Cross-Modal Retrieval"], ["wang2019semi", "Semi-supervised Deep Quantization for Cross-modal Search"], ["liu2020joint", "Joint-modal Distribution-based Similarity Hashing for Large-scale Unsupervised Deep Cross-modal Retrieval"], ["andoni2006near", "Unsupervised Hashing with Similarity Distribution Calibration"]]

publications-metadata/zhu2013linear.json

@@ -1 +1 @@
-[["chen2024supervised", "Supervised Consensus Anchor Graph Hashing for Cross Modal Retrieval"], ["luo2018fast", "Fast Scalable Supervised Hashing"], ["sun2022deep", "Deep Normalized Cross-Modal Hashing With Bi-Direction Relation Reasoning"], ["zhang2014latent", "Supervised Hashing with Latent Factor Models"]]
+[["chen2024supervised", "Supervised Consensus Anchor Graph Hashing for Cross Modal Retrieval"], ["luo2018fast", "Fast Scalable Supervised Hashing"], ["zhang2014latent", "Supervised Hashing with Latent Factor Models"], ["li2018scratch", "SCRATCH: A Scalable Discrete Matrix Factorization Hashing for Cross-Modal Retrieval"]]

publications/andoni2006near/index.html

@@ -50,11 +50,11 @@
 <meta property="og:url" content="https://learning2hash.github.io/publications/andoni2006near/" />
 <meta property="og:site_name" content="Awesome Learning to Hash" />
 <meta property="og:type" content="article" />
-<meta property="article:published_time" content="2024-05-22T05:47:34-05:00" />
+<meta property="article:published_time" content="2024-05-22T05:47:54-05:00" />
 <meta name="twitter:card" content="summary" />
 <meta property="twitter:title" content="Near-Optimal Hashing Algorithms for Approximate Nearest Neighbor in High Dimensions" />
 <script type="application/ld+json">
-{"@context":"https://schema.org","@type":"BlogPosting","dateModified":"2024-05-22T05:47:34-05:00","datePublished":"2024-05-22T05:47:34-05:00","description":"We present an algorithm for the c-approximate nearest neighbor problem in a d-dimensional Euclidean space, achieving query time of O(dn 1c2/+o(1)) and space O(dn + n1+1c2/+o(1)). This almost matches the lower bound for hashing-based algorithm recently obtained in (R. Motwani et al., 2006). We also obtain a space-efficient version of the algorithm, which uses dn+n logO(1) n space, with a query time of dnO(1/c2). Finally, we discuss practical variants of the algorithms that utilize fast bounded-distance decoders for the Leech lattice","headline":"Near-Optimal Hashing Algorithms for Approximate Nearest Neighbor in High Dimensions","mainEntityOfPage":{"@type":"WebPage","@id":"https://learning2hash.github.io/publications/andoni2006near/"},"url":"https://learning2hash.github.io/publications/andoni2006near/"}</script>
+{"@context":"https://schema.org","@type":"BlogPosting","dateModified":"2024-05-22T05:47:54-05:00","datePublished":"2024-05-22T05:47:54-05:00","description":"We present an algorithm for the c-approximate nearest neighbor problem in a d-dimensional Euclidean space, achieving query time of O(dn 1c2/+o(1)) and space O(dn + n1+1c2/+o(1)). This almost matches the lower bound for hashing-based algorithm recently obtained in (R. Motwani et al., 2006). We also obtain a space-efficient version of the algorithm, which uses dn+n logO(1) n space, with a query time of dnO(1/c2). Finally, we discuss practical variants of the algorithms that utilize fast bounded-distance decoders for the Leech lattice","headline":"Near-Optimal Hashing Algorithms for Approximate Nearest Neighbor in High Dimensions","mainEntityOfPage":{"@type":"WebPage","@id":"https://learning2hash.github.io/publications/andoni2006near/"},"url":"https://learning2hash.github.io/publications/andoni2006near/"}</script>
 <!-- End Jekyll SEO tag -->