A Framework for Automatic Burn Image Segmentation and Burn Depth Diagnosis Using Deep Learning

Figure 7

The segmentation network composed of ResNet-50/ResNet-101 and UPerNet. The four feature maps of ResNet-50/ResNet-101 on the left feature pyramid are the output of the 10th, 22nd, 40th and 49th/10th, 22nd, 91st, and 100th convolutional layers. The sizes of the four downsampling feature maps, respectively, are 1/4, 1/8, 1/16, and 1/32 of the original input image size.