Skip to main content
. 2024 Feb 14;14:3713. doi: 10.1038/s41598-024-54186-7

Figure 3.

Figure 3

Saliency maps for implicit concepts (left) and explicit concepts (right) learned by individual filters of the CNN model. It is interesting to note that there are no labels for implicit concepts in the training dataset. Warmer regions represent a high score for the specified concept in the prediction map. Note that EB and DB denote the encoder and decoder block layers, individually.