[Retracted] Channel-Wise Correlation Calibrates Attention Module for Convolutional Neural Networks

Figure 4

Score-CAM visualization results. We compare the visualization results of LC module-integrated network (ResNet50+LCM) with the results of the baseline (ResNet50) and SE-integrated network (ResNet50+SE). The Score-CAM visualization is calculated for the final convolutional outputs. On the middle of each input image, the ground-truth label is shown. And represents the softmax score of each network for the ground-truth class. On the right side of the input image, the first line represents score-weighted class activation heatmap and the second line represents score-weighted class activation heatmap on image. In the two color bars, the corresponding degree gradually decreases from top to bottom.