Figure 5
From: Insights and approaches using deep learning to classify wildlife
GG-CAM generated localized discriminative visual features of randomly selected images of baboon and impala. For classifying baboons, the CNN focuses on faces and tails. For impalas, the CNN uses the contrast between the white underbelly and dark back, black streaks on the rear, and black spots between the rear legs and underbelly. Most of the features extracted by the CNN have counterparts (similar focal visual components) in the human visual descriptors (indicated by the colors and agreed upon by at least 2 of 4 authors). The similarity is calculated as the DSC between extracted features and corresponding human descriptors (further detail in Fig. 8, Appendix 5).
