Understanding Intra-Class Knowledge Inside CNN

By Donglai Wei, Bolei Zhou, Antonio Torralba, William T. Freeman. 2015

94 citations recorded by the source corpus.

Convolutional Neural Network (CNN) has been successful in image recognition tasks, and recent works shed lights on how CNN separates different classes with the learned inter-class knowledge through visualization. In this work, we instead visualize the intra-class knowledge inside CNN to better understand how an object class is represented in the fully-connected layers. To invert the intra-class knowledge into more interpretable images, we propose a non-parametric patch prior upon previous CNN visualization models. With it, we show how different "styles" of templates for an object class are organized by CNN in terms of location and content, and represented in a hierarchical and ensemble way. Moreover, such intra-class knowledge can be used in many interesting applications, e.g. style-based image retrieval and style-based object completion.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Model Architectures

Citations in the corpus

What makes ImageNet good for transfer learning? (2016, background): “Another line of prior work investigated questions such as how distributed are CNN representations [2, 43], what is the the effect of feature transformations such as binarization [2, 14], how do CNN features change with systematic variations in scene factors [3], how invariant are features to differe”
Synthesizing the preferred inputs for neurons in neural networks via deep generator networks (2016, background): “Many handdesigned natural image priors have been experimentally shown to improve image quality such as: Gaussian blur [7], α-norm [5, 7, 8], total variation [6, 9], jitter [10, 6, 9], data-driven patch priors [8], center-bias regularization [9], and initializing from mean images [9].”
Towards Better Analysis of Deep Convolutional Neural Networks (2016, methodology): “Existing methods can be classified into two categories, namely, code inversion [12, 35, 58] and activation maximization [13, 41, 45, 53, 57].”
One Pixel Attack for Fooling Deep Neural Networks (2017, background): “Many efforts have been paid for understanding DNN by visualizing the activation of network nodes [15, 16, 17], on the otherside the geometrical characteristics of DNN boundary have gained less attraction due to the difficulty of understanding high-dimensional space.”
Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural Networks (2016, methodology): “Wei et al., 2015 proposed a more advanced data-driven patch prior regularization to enforce the visualizations to match the colors of a set of natural image patches.”
Do Convolutional Neural Networks Learn Class Hierarchy? (2017, background): “Accounting for the multifaceted nature of neurons [49, 71] and imposing natural image priors [43, 47] have significantly improved the interpretability.”
Diverse feature visualizations reveal invariances in early layers of deep neural networks (2018, unknown): “20] or the distance between the generated visualizations and natural images patches [32], or performing gradient descent in the feature space of a deep generator network [19].”
Definitions, methods, and applications in interpretable machine learning (2019, methodology): “When visualizing convolutional neural networks trained on image data, work has been done on visualizing ﬁlters (62, 63), maximally activating responses of individual neurons or classes (64), understanding intra-class variation (65), and grouping diﬀerent neurons (66).”
Visual Analytics in Deep Learning: An Interrogative Survey for the Next Frontiers (2018, unknown): “Regarding image generation and feature visualization, while some techniques have proven interesting [99], one of the most studied techniques, class activation maximization, maximizes the activation of a chosen, specific neuron using an optimization scheme, such as gradient ascent, and generates synt”
A Powerful Generative Model Using Random Weights for the Deep Image Representation (2016, background): “In addition, some regularizers are incorporated as a natural image prior to improve the visualization quality, including α−norm [6], total variation [7], jitter [7], Gaussian blur [9], data-driven patch priors [17], etc.”