Neural Codes for Image Retrieval

By David Stutz. 2015

876 citations recorded by the source corpus.

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

PlaNet - Photo Geolocation with Convolutional Neural Networks (2016, methodology): “Their features even out-perform ours, which is likely because [3] use a carefully crafted landmark dataset for re-training while we applied only minimal ﬁltering.”
Grafit: Learning fine-grained image representations with coarse labels (2020, methodology): “For category-level retrieval we report the mean average precision (mAP), as commonly done for retrieval tasks [2, 42].”
Unsupervised Domain Adaptation by Backpropagation (2014, methodology): “In the context of deep feed-forward architectures, such data can be used to “ﬁne-tune” the network trained on the source domain [24, 15, 1].”
Domain-Adversarial Training of Neural Networks (2015, methodology): “In the context of deep feed-forward architectures, such data can be used to “ﬁne-tune” the network trained on the source domain (Zeiler and Fergus, 2013; Oquab et al., 2014; Babenko et al., 2014).”
MultiGrain: a unified image embedding for classes and instances (2019, methodology): “After the emergence of convolutional neural networks (CNNs) for large-scale classiﬁcation on ImageNet [28, 37], it has become apparent that CNNs trained on classiﬁcation datasets are very competitive image feature extractors for various vision tasks, including instance retrieval [1, 9, 35].”
Human Action Recognition Using Factorized Spatio-Temporal Convolutional Networks (2015, background): “What is more, such a factorized scheme enables the learning of 2D spatial kernels to benefit from the existing massive image datasets [3] with which the performance of many vision tasks [4, 19, 1] have been boosted significantly.”
Particular object retrieval with integral max-pooling of CNN activations (2015, methodology): “Several works have already investigated this research direction, such as global or local representations based on either fully connected (Babenko et al., 2014; Gong et al., 2014) or convolutional layers (Razavian et al., 2014b; Azizpour et al., 2014; Babenko & Lempitsky, 2015).”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “The second and most used solution is to ﬁne-tune the deep network model on the new type of data and for the new task [268, 157, 9, 43].”
End-to-End Learning of Deep Visual Representations for Image Retrieval (2016, methodology): “We leverage a large-scale image dataset, the Landmarks dataset Babenko et al (2014), that contains approximately 214k images of 672 famous landmark sites.”
Fisher vectors meet Neural Networks: A hybrid classification architecture (2015, background): “This finding was quantitatively validated for a number of tasks including image classification [13, 34, 56, 6, 41], image retrieval [41, 1], object detection [16], and action recognition [34].”
Polysemous Codes (2016, methodology): “We extract convolutional neural networks features [35] following the guidelines of [9]: we compute the activations of the 7 th layer of AlexNet [33].”
Learning With Average Precision: Training Image Retrieval With a Listwise Loss (2019, methodology): “Compared to classification losses previously used for retrieval with less success [3, 4, 47], ranking-based loss functions directly optimize for the end task, enforcing intraclass discrimination and more fine-grained instance-level image representations [18].”
Dynamic Match Kernel With Deep Convolutional Features for Image Retrieval (2018, background): “[41] extensively evaluate the performance of deep features.”
Siamese network features for image matching (2016, positive): “Like our approach, both methods [3], [12] are based on CNN but using different objective and network structures which are not able to handle image pairs directly.”
CNN Image Retrieval Learns from BoW: Unsupervised Fine-Tuning with Hard Examples (2016, methodology): “[14] go one step further and re-train such networks with a dataset that is closer to the target task.”
Large-Scale Image Retrieval with Attentive Deep Local Features (2016, methodology): “In the past few years, several global descriptors based on CNNs have been proposed to use pretrained [4, 34] or learned networks [2, 29, 11].”
Google Landmarks Dataset v2 – A Large-Scale Benchmark for Instance-Level Recognition and Retrieval (2020, background): “7M StreetView City Y Landmarks-PointCloud [35] 2012 1k 10k 205k Flickr label + SfM Worldwide Y 24/7 Tokyo [55] 2015 125 315 1k Smartphone + Manual City Y Paris500k [60] 2015 13k 3k 501k Manual City N Landmark URLs [7, 22] 2016 586 140k - Text query + Feature matching Worldwide N Flickr-SfM [44] 2016”
Cross-Dimensional Weighting for Aggregated Deep Convolutional Features (2015, background): “Algorithm 1 summarises these steps as pseudocode.”
DeepHash: Getting Regularization, Depth and Fine-Tuning Right (2015, methodology): “For the proposed deep Siamese ﬁne-tuning scheme proposed, we use the 200K matching image pairs data set provided by Yandex in their recent work [3], consisting primarily of landmark images.”