Locally Optimized Product Quantization for Approximate Nearest Neighbor Search

Kalantidis, Yannis; Avrithis, Yannis

doi:10.1109/cvpr.2014.298

Cited by 227 publications

(213 citation statements)

References 15 publications

Supporting

Mentioning

210

Contrasting

Unclassified

Order By: Relevance

“…Betimleyici eşlemesi için son zamanlarda yaygınlaşan iki yaklaşım Çarpım Nicemlemesi (Product Quantization) [24] ve çizge üzerinde yaklaşık en yakın komşu hesaplanmasıdır [25]. Ancak bu yöntemler henüz ikilik betimleyicilere uyarlanmadığından gerçek zamanlı mobil uygulamalarda kullanılmaları mümkün değildir.…”

Section: Yaklaşık En Yakın Komşu Yöntemleriunclassified

Object detection with BRIEF descriptors and locality sensitive matching for augmented reality

Özuysal¹

2017

Pamukkale J Eng Sci

View full text Add to dashboard Cite

Section: Yaklaşık En Yakın Komşu Yöntemleriunclassified

Object detection with BRIEF descriptors and locality sensitive matching for augmented reality

Özuysal¹

2017

Pamukkale J Eng Sci

View full text Add to dashboard Cite

“…Additive quantization even with pyramid encoding is still significantly slower than PQ or OPQ; therefore, a significant search performance benefit is required to justify the approach. When orthogonal initialization is used, the suitable number of PQ iterations to run was experimentally found to be around [10][11][12][13][14][15]. Thus the extra initialization costs do not affect the asymptotic complexity estimate.…”

Section: Pyramid Encodingmentioning

confidence: 99%

“…The computational complexity of learning an OPQ quantizer is ( ( + + )) in the non-parametric case. Other suggested variants of product quantization include Locally Optimized Product Quantization (LOPQ) [13] and Optimized Cartesian K-Means (OCKM) [14].…”

Section: Introductionmentioning

confidence: 99%

Pyramid encoding for fast additive quantization

Muravev

Ozan

Iosifidis

et al. 2017

2017 25th European Signal Processing Conference (EUSIPCO)

View full text Add to dashboard Cite

Abstract-The problem of approximate nearest neighbor (ANN) search in Big Data has been tackled with a variety of recent methods. Vector quantization based solutions have been maintaining the dominant position, as they operate in the original data space, better preserving inter-point distances. Additive quantization (AQ) in particular has pushed the state-of-the-art in search accuracy, but high computational costs of encoding discourage the practical application of the method. This paper proposes pyramid encoding, a novel technique, which can replace the original beam search to provide a significant complexity reduction at the cost of a slight decrease in retrieval performance. AQ with pyramid encoding is experimentally shown to obtain results comparable with the baseline method in accuracy, while offering significant computational benefits.

show abstract

“…Jégou et al [18] proposed an index that divides the space into a set of Voronoi cells through kmeans based vector quantization. Further works improve candidate distance computation [19], descriptor quantization [21] and more e ective centroid evaluation [5]. Tavenard et al [33] proposed a technique for balancing k-means cluster size, by shi ing cluster boundaries into parallel boundaries.…”

Section: Related Workmentioning

confidence: 99%

Balanced Search Space Partitioning for Distributed Media Redundant Indexing

Mourão

Magalhães

2017

Proceedings of the 2017 ACM on International Conference on Multimedia Retrieval

View full text Add to dashboard Cite

is paper addresses the problem of balanced, redundant indexing of media information. Our goal is to partition and distribute the search index, taking advantage of the distributed systems properties: balanced load across nodes, redundancy on node down and e cient node usage under concurrent querying. We follow an information compression approach to solve this problem and propose to represent data with overcomplete codebooks, where each document is represented by only a few codewords and an indexing node is responsible for several codewords.antization algorithms are designed to t the original data as best as possible, leading to bias towards codewords that t the principal directions of data. In this paper, we propose the balanced KSVD (B-KSVD) algorithm, that distributes the allocation of data across a balanced number of codewords, according to the global distribution of data. Indexing experiments showed that B-KSVD can achieve 38% 1-recall by inspecting only 1% of the full index, distributed over 10 partitions. Traditional methods based on k-means need to either use larger codebooks or to inspect a larger portion of the index to achieve the same retrieval performance.

show abstract

Locally Optimized Product Quantization for Approximate Nearest Neighbor Search

Cited by 227 publications

References 15 publications

Object detection with BRIEF descriptors and locality sensitive matching for augmented reality

Object detection with BRIEF descriptors and locality sensitive matching for augmented reality

Pyramid encoding for fast additive quantization

Balanced Search Space Partitioning for Distributed Media Redundant Indexing

Contact Info

Product

Resources

About