ePrints@IIScePrints@IISc Home | About | Browse | Latest Additions | Advanced Search | Contact | Help

Saliency-Driven Class Impressions for Feature Visualization of Deep Neural Networks

Addepalli, S and Tamboli, D and Babu, RV and Banerjee, B (2020) Saliency-Driven Class Impressions for Feature Visualization of Deep Neural Networks. In: Proceedings - International Conference on Image Processing, 25-28, September 2020, Virtual, Abu Dhabi; United Arab Emirates, pp. 1936-1940.

Pro-020S.pdf - Published Version

Download (259kB) | Preview
Official URL: https://dx.doi.org/10.1109/ICIP40778.2020.9190826


In this paper, we propose a data-free method of extracting Impressions of each class from the classifier's memory. The Deep Learning regime empowers classifiers to extract distinct patterns (or features) of a given class from training data, which is the basis on which they generalize to unseen data. Before deploying these models on critical applications, it is very useful to visualize the features considered to be important for classification. Existing visualization methods develop high confidence images consisting of both background and foreground features. This makes it hard to judge what the important features of a given class are. In this work, we propose a saliency-driven approach to visualize discriminative features that are considered most important for a given task. Another drawback of existing methods is that, confidence of the generated visualizations is increased by creating multiple instances of the given class. We restrict the algorithm to develop a single object per image, which helps further in extracting features of high confidence, and also results in better visualizations. We further demonstrate the generation of negative images as naturally fused images of two or more classes. Our code is available at: https://giChub.com/val-iisc/Saliency-driven-Class-Impressions. © 2020 IEEE.

Item Type: Conference Paper
Publication: Proceedings - International Conference on Image Processing, ICIP
Publisher: IEEE Computer Society
Additional Information: cited By 0; Conference of 2020 IEEE International Conference on Image Processing, ICIP 2020 ; Conference Date: 25 September 2020 Through 28 September 2020; Conference Code:165772
Keywords: Classification (of information); Data mining; Deep learning; Deep neural networks; Neural networks; Visualization, Critical applications; Discriminative features; Extracting features; Foreground features; High confidence; Important features; Multiple instances; Visualization method, Image processing
Department/Centre: Division of Interdisciplinary Sciences > Computational and Data Sciences
Date Deposited: 22 Jan 2021 06:39
Last Modified: 22 Jan 2021 06:39
URI: http://eprints.iisc.ac.in/id/eprint/67728

Actions (login required)

View Item View Item