Figure 3.

Chart comparing performance of three tested CNNs, explaining the selection of the backbone feature extraction network in the Dilated ResFCN architecture. Number of operations corresponds to a single forward pass of the backbone network. The number of parameters for each network is represented by the corresponding circle [47]. The values shown as red crosses were obtained using Dilated ResFCN architecture with ResNet50, ResNet101, and ResNet152 as feature extraction backbone networks (shown as blue in Figure 2), with the corresponding standard deviation estimates represented by the blue vertical bars.