Author Manuscript, Published in "ieee Conference on Computer Vision and Pattern Recognition on the Burstiness of Visual Elements

By Herv'e J'egou, Matthijs Douze, Cordelia Schmid

Subfields: Computer Vision and Pattern Recognition

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

Aggregating Local Image Descriptors into Compact Codes (2012, background): “First, it reduces the influence of bursty visual elements, which were shown to corrupt the image similarity in [29].”
Return of the Devil in the Details: Delving Deep into Convolutional Nets (2014, methodology): “In this case, the 2 normalisation is applied to the individual sub-blocks ( u k , v k ) of the vector φ FV ( I ) , which helps to alleviate the local feature burstiness [28].”
NetVLAD: CNN Architecture for Weakly Supervised Place Recognition (2015, neutral): “While there have been many improvements in designing better image retrieval [22], [23], [24], [25], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], [62] and place recognition [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16] systems,”
Tensor Representations via Kernel Linearization for Action Recognition from 3D Skeletons (2016, background): “An example of such a commonly encountered property is the so-called burstiness [7], which is the property that a given feature appears more often in a sequence than a statistically independent model would predict.”
Dynamic Match Kernel With Deep Convolutional Features for Image Retrieval (2018, unknown): “[64] [51] [1] [26] [18] [53] [65] [66] [12] [52] [67] [58] [62] [19]”
Low-Rank Bilinear Pooling for Fine-Grained Classification (2016, result): “Our observation is consistent with the literature on applying normalization to deal with visual burstiness [11, 20].”
Multi-scale pyramid pooling for deep convolutional representation (2015, methodology): “Following the Improved Fisher Kernel framework [27], we finally apply power normalization to tackle burstiness [19, 14] and l2normalization to the Fisher vector G .”
All About VLAD (2013, methodology): “clustering on SIFTs): (i) Paris 6k [14], which is analogous to the Oxford buildings dataset, and is often used as an independent dataset from the Oxford buildings [1, 3, 5, 14]; (ii) Flickr60k [6], which contains 60k images downloaded from Flickr, and is used as an independent dataset from the Holid”
Learning a Fine Vocabulary (2010, unknown): “[12, 24] report the mAP of 0.”
Visual Place Recognition with Repetitive Structures (2015, methodology): “[13] study the effect of visual “burstiness”, i.”
Visual landmark recognition from Internet photo collections: A large-scale evaluation (2014, background): “improved feature representation [55], improved feature quantization [18, 56], or ranking methods robust to problems like repeated patterns [57] (Fig.”
COSTA: Co-Occurrence Statistics for Zero-Shot Classification (2014, methodology): “The square-root function is used in image retrieval and classification to correct for burstiness of visual features [10, 24].”
Cross-Dimensional Weighting for Aggregated Deep Convolutional Features (2015, background): “This implies some spatially recurring visual elements in the image, that were shown to negatively affect matching [13].”
Few-Shot Learning via Saliency-Guided Hallucination of Samples (2019, background): “Second-order statistics have to deal with the so-called burstiness which is “the property that a given visual element appears more times in an image than a statistically independent model would predict” [14].”
Packing bag-of-features (2009, background): “Searching for images of the same object or scene in a large number of images has recently received increasing attention [9, 4, 11, 5].”
Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking (2018, methodology): “The codebook size is 65,536 in our experiments, while burstiness normalization [17] is always used.”
Visual Word Ambiguity (2010, background): “The authors weight closer neighbors heavier than farther ones, without normalizing the scores to a posterior probability.”
Learning Representations by Predicting Bags of Visual Words (2020, methodology): “In terms of effect over the weight vector, this reparametrization is similar with the power-law normalization [53] used for mitigating the burstiness effect on BoW-like representations [31].”