Distinct Class Saliency Maps for Multiple Object Images

Wataru Shimoda, Keiji Yanai

Abstract:

This paper proposes a method to obtain more distinct class saliency maps than
Simonyan et al. (2014). We made three improvements over their method: (1) using
CNN derivatives with respect to feature maps of the intermediate convolutional
layers with up-sampling instead of an input image; (2) subtracting saliency
maps of the other classes from saliency maps of the target class to differentiate
target objects from other objects; (3) aggregating multi-scale class saliency maps
to compensate lower resolution of the feature maps.

Conflicts: uec.ac.jp