Skip to main content
. 2021 Jan 16;113:107828. doi: 10.1016/j.patcog.2021.107828

Fig. 4.

Fig. 4

A brief illustration of the learning principle for the proposed global contrast pooling (GCP) layer. Here, the concepts denote to-be-learned features that are discriminative for severity assessment. The GCP layer is designed to pull the relevant instance features and concepts closer, and push the irrelevant instance features and concepts away from each other.