Command Palette
Search for a command to run...
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
Simonyan Karen Vedaldi Andrea Zisserman Andrew
Abstract
This paper addresses the visualisation of image classification models, learntusing deep Convolutional Networks (ConvNets). We consider two visualisationtechniques, based on computing the gradient of the class score with respect tothe input image. The first one generates an image, which maximises the classscore [Erhan et al., 2009], thus visualising the notion of the class, capturedby a ConvNet. The second technique computes a class saliency map, specific to agiven image and class. We show that such maps can be employed for weaklysupervised object segmentation using classification ConvNets. Finally, weestablish the connection between the gradient-based ConvNet visualisationmethods and deconvolutional networks [Zeiler et al., 2013].
Code Repositories
Benchmarks
| Benchmark | Methodology | Metrics |
|---|---|---|
| image-attribution-on-celeba | Saliency | Deletion AUC score (ArcFace ResNet-101): 0.1453 Insertion AUC score (ArcFace ResNet-101): 0.4632 |
| image-attribution-on-cub-200-2011-1 | Saliency | Deletion AUC score (ResNet-101): 0.0682 Insertion AUC score (ResNet-101): 0.6585 |
| image-attribution-on-vggface2 | Saliency | Deletion AUC score (ArcFace ResNet-101): 0.1907 Insertion AUC score (ArcFace ResNet-101): 0.5612 |
| interpretability-techniques-for-deep-learning-1 | Saliency | Insertion AUC score: 0.4632 |
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.