Using very deep autoencoders for content-based image retrieval

By Alex Krizhevsky, Geoffrey E. Hinton. 2011

427 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

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 Representation & Self-Supervised Learning

Citations in the corpus

A Retrieve-and-Edit Framework for Predicting Structured Outputs (2018, background): “Semantic hashing by autoencoders [16] is a related idea where an autoencoder’s latent representation is used to construct a hash function to identify similar images or texts [29, 6].”
Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations (2017, background): “Among the most popular ones are stochastic approximations [39, 19, 7, 32, 5] and rounding with a smooth derivative approximation [15, 30].”
Full Resolution Image Compression with Recurrent Neural Networks (2016, methodology): “Autoencoders have been used to reduce the dimensionality of images [9], convert images to compressed binary codes for retrieval [13], and to extract compact visual representations that can be used in other applications [18].”
Variable Rate Image Compression with Recurrent Neural Networks (2015, positive): “For some types of autoencoder, encoding the bottleneck as a simple bit vector can be beneﬁcial (Krizhevsky & Hinton, 2011).”
Joint Unsupervised Learning of Deep Representations and Image Clusters (2016, background): “One class of approaches cater to reconstruction tasks, such as autoencoders [21, 28, 33, 41, 53], deep belief networks (DBN) [31], etc.”
An Introduction to Deep Learning (2011, positive): “A similar approach for a large scale image database is presented in this special session [41].”
Audio-visual speech recognition using deep learning (2014, methodology): “Parameters for the network structures are empirically determined with reference to previous studies [18, 21].”
A Review of Deep Learning Methods and Applications for Unmanned Aerial Vehicles (2017, background): “In this process, the weights initialization is critical to avoid reaching a bad local optimum; thus some authors have proposed a pretrained stage based on stacked RBMs and a fine-tuning stage using backpropagation [24, 27].”
Improved Lossy Image Compression with Priming and Spatially Adaptive Bit Rates for Recurrent Networks (2017, methodology): “Autoencoders with a bottleneck have been used to learn compact representations for many applications [11, 16, 24] and form the basis for most network-based compression models.”
Multimodal integration learning of robot behavior using deep neural networks (2014, neutral): “The parameters for the network structures are empirically determined in reference to such previous studies as [27] and [33].”
Pathology Image Analysis Using Segmentation Deep Learning Algorithms. (2019, background): “…which aims to reconstruct the inputs, can then be used to train the network (referred to as unsupervised pre-training) before supervised training 59,60 ; because no ground truth labeling is required in this step, this pre-training strategy helps to improve model performance by utilizing…”
Deep Asymmetric Transfer Network for Unbalanced Domain Adaptation (2018, methodology): “Nowadays, deep learning models have been successfully applied to many applications (Krizhevsky and Hinton 2011; Mikolov et al. 2010; Wang, Cui, and Zhu 2016).”
Image compression with Stochastic Winner-Take-All Auto-Encoder (2017, neutral): “However, two issues have so far prevented them from becoming efficient image compression algorithms: they can only be trained for one image size and one compression rate [2, 3].”
Autoencoding the retrieval relevance of medical images (2015, methodology): “Moreover, to solve the spatial information problem in a histogram, the color coherence vectors (CCV) method has been proposed [12].”
Autoencoder Based Image Compression: Can the Learning be Quantization Independent? (2018, neutral): “If an autoencoder has fully-connected layers [10, 11, 12], the number of parameters depends on the image size.”
Hashing with binary autoencoders (2015, methodology): “lying a threshold to the encoder (Salakhutdinov and Hinton, 2009), or rounding the encoder output to 0 or 1 during the backpropagation forward pass but ignoring the rounding during the backward pass (Krizhevsky and Hinton, 2011). 2 3 Our hashing models: binary autoencoder and binary factor analysis ”