Figure 2:
Neural network architecture (based on TernausNet). The feature maps marked in yellow represents the VGG16 encoder pre-trained with the ImageNet dataset. The feature maps in red represent the decoder which is trained using the fluoroscopy images.