Aggregating Local Deep Features for Image Retrieval

By Artem Babenko, Victor S. Lempitsky. 2015

197 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Information Retrieval & Search

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

In Media Technology

In Information Retrieval & Search

Citations in the corpus

NetVLAD: CNN Architecture for Weakly Supervised Place Recognition (2015, methodology): “This confirms two important premises of this work: (i) our approach can learn rich yet compact image representations for place recognition, and (ii) the popular idea of using pretrained networks “off-the-shelf” [7, 8, 23, 60, 61] is sub-optimal as the networks trained for object or scene classificat”
Deep Feature Factorization for Content-Based Image Retrieval and Localization (2019, unknown): “, global max-pooling [4] (MAC, maximum activation of convolutions), sum-pooling [5], and weighted-sum-pooling [6].”
Google Landmarks Dataset v2 – A Large-Scale Benchmark for Instance-Level Recognition and Retrieval (2020, methodology): “These were learned with different backbones such as ResNet [25], ResNeXt [62], Squeeze-andExcitation [27], FishNet [52] and Inception-V4 [53]; pooling methods such as SPoC [6], RMAC [55] or GeM [45]; loss functions such as ArcFace [18], CosFace [58], N-pairs [50] and triplet [49].”
SIFT Meets CNN: A Decade Survey of Instance Retrieval (2016, methodology): “We also note in this survey that several other works [133], [134], [140] also employ similar ideas with [10] in employing max or average pooling on the intermediate feature maps and that Razavian et al.”
Convolutional Neural Network Architecture for Geometric Matching (2017, methodology): “produces a feature map f 2Rh w d, which can be interpreted as a h wdense spatial grid of d-dimensional local descriptors. A similar interpretation has been used previously in instance retrieval works [3,4,5,16] demonstrating high discriminative power of CNN-based descriptors. Thus, for feature extra”
Unifying Deep Local and Global Features for Image Search (2020, methodology): “in computer vision, they were developed mainly by aggregating hand-crafted local descriptors [25,26,27,58]. Today, most high-performing global features are based on deep convolutional neural networks [5,4,59,1,17,47,48], which are trained with ranking-based [9,51,19] or classiﬁcation losses [61,11].”
Large-Scale Image Retrieval with Attentive Deep Local Features (2016, methodology): “nd-crafted features in conventional aggregation techniques such as VLAD or FV [24, 36]. Other works have re-evaluated and proposed different feature aggregation methods using such deep local features [3,21]. CNNs have also been used to detect, represent and compare local image features. Verdie et al”
Hierarchical Bilinear Pooling for Fine-Grained Visual Recognition (2018, background): “ subtle and local, capturing global semantic information with merely fully connected layers limits the representation capacity of a framework, and hence restricts further promotion of nal recognition [1]. To better model subtle dierence for ne-grained categories, Lin et al. [17] proposed a bilinear ”
A Decade Survey of Content Based Image Retrieval Using Deep Learning (2020, methodology): “have applied the aggregation of deep convolutional features to form the discriminative descriptors for image retrieval [194].”