Principal Feature Visualisation in Convolutional Neural Networks
Peer reviewed, Journal article
Accepted version
View/ Open
Date
2020Metadata
Show full item recordCollections
- Publikasjoner fra CRIStin - SINTEF AS [5668]
- SINTEF Digital [2415]
Original version
Lecture Notes in Computer Science (LNCS). 2020, 12368, 18-31. 10.1007/978-3-030-58592-1_2Abstract
We introduce a new visualisation technique for CNNs called Principal Feature Visualisation (PFV). It uses a single forward pass of the original network to map principal features from the final convolutional layer to the original image space as RGB channels. By working on a batch of images we can extract contrasting features, not just the most dominant ones with respect to the classification. This allows us to differentiate between several features in one image in an unsupervised manner. This enables us to assess the feasibility of transfer learning and to debug a pre-trained classifier by localising misleading or missing features.