ИСТИНА

Saliency map highlights portion of the image that contributes to a classification decision, hence it is an effective tool for the interpretation of convolutional neural networks (CNN). In recent years, various saliency detection methods have been proposed: Guided Backpropagation(GBP), Class Activation Mapping(CAM), Grad-CAM, Grad-CAM++ and so on. Although some saliency detection methods are able to generate the saliency map, they are independent of the model and data generation process, and may not best explain the relationship between the inputs and outputs of the model during learning or to debug the model. Inspired by the lateral inhibition(LI) mechanism in the brain, we introduced it in artificial neural networks and applied it to saliency detection. Experiments show that our approach has significant advantages in comparison with the mainstream methods.