On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation (original) (raw)
Fig 27
The pixel-wise decomposition is different from an edge or texture detector.
Only a subset of strong edges and textures receive high scores. Panels show the original image on the left, and the decomposition on the right. The decompositions were computed twice for the classes table lamp and once for the class rooster. The neural net is the pre-trained one on ILSVRC data from the Caffe package [60]. The computing methods for each column are the same as in Fig 25. The last row shows the gradient norms normalized to lie in [0, 1] mapped by the same color scheme as for the heatmaps. Pictures in order of appearance from Wikimedia Commons by authors Wtshymanski, Serge Ninanne and Immanuel Clio.