Fig. 5: Architecture diagram of the Back High-Resolution Network outlining the deep learning model architecture. | BDJ Open

Fig. 5: Architecture diagram of the Back High-Resolution Network outlining the deep learning model architecture.

From: A rapid identification method for soft tissue markers of dentofacial deformities based on heatmap regression

Fig. 5

The architecture has 4 stages. Stage 1 obtains the branches of feature maps with different resolutions. Stage 2 to Stage 4 obtain and fuse multidimensional feature maps. Using the opposite network structure of HR-Net, the input image is downsampled several times to obtain features of different sizes, which are then fused; at the same time, upsampling ensures that the minimum size features are fully utilized to obtain richer semantic information.

Back to article page