Fig. 3From: Detecting representative characteristics of different genders using intraoral photographs: a deep learning model with interpretation of gradient-weighted class activation mappingRepresentative Gaussian heatmaps in 7–14 yrs female group demonstrating area of interest, which were falsely predicted by our network. Figure a–c showed images where mistakes were corrected by cropping. Figure d–e showed images where mistakes occurred after being cropped. Figure f–g showed images which were incorrectly labelled both before and after being croppedBack to article page