Neural Codes for Image Retrieval

By Artem Babenko, Anton Slesarev, Alexander Chigorin, Victor S. Lempitsky. European Conference on Computer Vision, 2014

392 citations recorded by the source corpus.

It has been shown that the activations invoked by an image within the top layers of a large convolutional neural network provide a high-level descriptor of the visual content of the image. In this paper, we investigate the use of such descriptors (neural codes) within the image retrieval application. In the experiments with several standard retrieval benchmarks, we establish that neural codes perform competitively even when the convolutional neural network has been trained for an unrelated classification task (e.g. Image-Net). We also evaluate the improvement in the retrieval performance of neural codes, when the network is retrained on a dataset of images that are similar to images encountered at test time.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5c3dce55c61ee86073575ac75cc882a215cb49e6

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep Feature Factorization for Content-Based Image Retrieval and Localization (2019, methodology): “Early approaches to CNN-based image retrieval made use of activations of CNN fully-connected layers [2, 3], since they naturally provided a fixed-size global image descriptor.”
“Zhores” — Petaflops supercomputer for data-driven modeling, machine learning and artificial intelligence installed in Skolkovo Institute of Science and Technology (2019, background): “nary research in data-driven simulations and modeling, machine learning, Big Data and articial intelligence (AI). It enables research in such important elds as Bio-medicine [46, 44], Computer Vision [20, 21, 11, 42, 45, 8], Remote Sensing Preprint submitted to Mathematics and Computers in Simulation”
Wafer Map Defect Pattern Classification and Image Retrieval Using Convolutional Neural Network (2018, background): “Since the deep CNN can learn rich features at each layer, these intermediate features are used as good descriptors for image retrieval [8], [9].”
DeepStore: In-Storage Acceleration for Intelligent Queries (2019, background): “Compared with the traditional key-based query systems that rely on structured data, intelligent query systems extract feature vectors from unstructured data using deep learning techniques [17, 24, 66].”
NetVLAD: CNN Architecture for Weakly Supervised Place Recognition (2015, background): “These include treating activations from certain layers directly as descriptors by concatenating them [9, 60], or by pooling [7, 8, 23].”
Good Practice in CNN Feature Transfer (2016, methodology): “In image search, CNN can be used as global [1, 41, 32], regional [8, 35], or local features [24].”
Learning visual similarity for product design with convolutional neural networks (2015, methodology): “2012] and extract the output from the 7th layer (fc7), which has been shown to be a high quality image descriptor [Babenko et al. 2014].”
A Decade Survey of Content Based Image Retrieval Using Deep Learning (2020, methodology): “6: The illustration of the neural code generation from a convolutional neural network (CNN) [67].”
Deep learning of binary hash codes for fast image retrieval (2015, methodology): “[1] focus on dimensional reduction of the CNN features, and improve the retrieval performance with compressed CNN features.”
Visual Instance Retrieval with Deep Convolutional Networks (2014, background): “The current paper is most related to 11)13)14)16) concerning the usage of ConvNet representations.”
Deep learning for visual understanding: A review (2016, methodology): “Finally, without sliding windows or multiple-scale patches, Babenko et al. [172] focus on holistic descriptors where the whole image is mapped to a single vector with a CNN model.”
Unifying Deep Local and Global Features for Image Search (2020, positive): “computer vision, they were developed mainly by aggregating hand-crafted local descriptors [25, 26, 27, 54]. Today, most high-performing global features are based on deep convolutional neural networks [5, 4, 55, 1, 17, 44, 45], which are trained with ranking-based [9, 48, 19] or classiﬁcation losses ”
Convolutional Neural Network-based Place Recognition (2014, positive): “In [Babenko, et al., 2014], mid-level features from CNNs are evaluated for the image retrieval application and achieve performance comparable to others using state-of-the-art features.”
Do We Train on Test Data? Purging CIFAR of Near-Duplicates (2019, methodology): “is not sufﬁcient, since there may also be slightly modiﬁed variants of the same scene that vary by contrast, hue, translation, stretching etc. Thus, we follow a content-based image retrieval approach [18, 2, 1] for ﬁnding duplicate and near-duplicate images: We train a lightweight CNN architecture p”
Deep exemplar-based colorization (2018, methodology): “To measure semantic similarity, we adopt the intermediate features of a pre-trained image classification network as descriptors, which have been widely used in recent image retrieval works [Krizhevsky et al. 2012; Babenko et al. 2014; Gong et al. 2014; Babenko and Lempitsky 2015; Razavian et al. 201”
Aggregating Local Deep Features for Image Retrieval (2015, methodology): “Several recent works [2, 21, 7] proposed to use the outputs of last fully-connected network layers as global image descriptors and demonstrate their advantage over prior state-of-the-art when the dimensionality of descriptors is limited.”