Separating Style and Content with Bilinear Models

By Josh Tenenbaum, William T. Freeman. 2000

953 citations recorded by the source corpus.

Subfields: Representation & Self-Supervised Learning, Signal Processing

How researchers see this paper

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

In Representation & Self-Supervised Learning

In Signal Processing

Citations in the corpus

InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets (2016, background): “One class of such methods trains a subset of the representation to match the supplied label using supervised learning: bilinear models [18] separate style and content; multi-view perceptron [19] separate face identity and view point; and Yang et al.”
A Neural Algorithm of Artistic Style (2015, background): “12,13 In our demonstration, we render a given photograph in the style of a range of well-known artworks.”
Generative Image Modeling Using Style and Structure Adversarial Networks (2016, background): “Finally, the idea of factorizing image into two separate phenomena has been well studied in [46,47,48,49], which motivates us to decompose the generative process to structure and style.”
Are Disentangled Representations Helpful for Abstract Visual Reasoning? (2019, background): “Disentangling pose, style, or motion from content are classical vision tasks that has been studied with different degrees of supervision [72, 79, 80, 34, 19, 14, 21, 36].”
Learning Deep Disentangled Embeddings with the F-Statistic Loss (2018, methodology): “With this compromise, embeddings can be used to model hierarchical class structure and can facilitate decomposing the instances according to underlying factors, e.g., separating content and style Tenenbaum & Freeman (2000).”
Disentangling Factors of Variation via Generative Entangling (2012, methodology): “Formally, the elements of observation x are given by x k = i j W ijk y i z j , ∀ k , where y i and z j are elements of the two factors ( y and z ) representing the observation and W ijk is an element of the tensor of model parameters (Tenenbaum and Freeman, 2000).”
Weakly-supervised Disentangling with Recurrent Transformations for 3D View Synthesis (2015, methodology): “Bilinear models for separating style and content are developed in [25], and are shown to”
Bilinear CNN Models for Fine-Grained Visual Recognition (2015, methodology): “Bilinear models were proposed by Tanenbaum and Freeman [33] to model two-factor variations, such as “style” and “content”, for images.”
Discovering Hidden Factors of Variation in Deep Networks (2014, methodology): “Previous approaches to separating factors of variation in data, such as content vs. style [3] or form vs. motion [4, 5, 6, 7, 8, 9], have relied upon a bilinear model architecture in which the units representing different factors are combined multiplicatively.”
Compact Bilinear Pooling (2015, methodology): “Bilinear models were first introduced by Tenenbaum and Freeman [32] to separate style and content.”
Deep Learning of Representations: Looking Forward (2013, methodology): “…priors on the form of the model can be used to enhance disentangling, such as multiplicative interactions between the factors (Tenenbaum and Freeman, 2000; Desjardins et al., 2012) or orthogonality of the features derivative with respect to the input (Rifai et al., 2011b, 2012a; Sohn et…”
A Survey of Inductive Biases for Factorial Representation-Learning (2016, background): “In a bilinear model [53], two factors interact with each other to generate each observation.”
Discovering states and transformations in image collections (2015, neutral): “This approach is reminiscent of Freeman and Tenenbaum’s work on separating style and content [29].”
Learning to Relate Images (2011, methodology): “Like an auto-encoder, this model can be trained with gradient based optimization.”
Joint Feature Selection and Subspace Learning for Cross-Modal Retrieval (2016, methodology): “Recently, several approaches for establishing inter-modal relationships between data from different modalities generally rely on subspace learning, such as Canonical Correlation Analysis [25], [26], Partial Least Squares (PLS) [27] and Bilinear Model (BLM) [28], [29].”