A NOVEL CLASS ACTIVATION MAP FOR VISUAL EXPLANATIONS IN MULTI-OBJECT SCENES

Class activation maps (CAMs) have emerged as a popular technique to improve model interpretability of deep learning-based models. While existing CAM methods are able to extract salient semantic regions to provide high-confidence pseudo-labels for downstream tasks such as semantic segmentation, they are less effective when dealing with multiple objects for CAM generation. In this paper, we design a multi-channel weight assignment scheme that learns from both positive and negative regions to improve the CAM for images with multiple objects. To demonstrate the effectiveness of our proposed method, we introduce two new data sets, a cat-and-dog dataset and a PASCAL VOC 2012-based multi-object dataset, to evaluate our approach and compare it with state-of-the-art CAM methods. The obtained results demonstrate that our CAM outperforms other approaches in terms of both mIoU and inter-object activation ratio, a new evaluation measure proposed to evaluate CAM performance in multi-object scenes.