Figure 3
From: High-throughput segmentation of unmyelinated axons by deep learning
U-Net segmentation model on tiles of size \(512 \times 512\) pixels, with 4 down-sampling (encoder) and up-sampling (decoder) blocks linked by skip connections. A batch normalization layer is inserted before the ReLU non-linearity in all the convolutional layers and dropout adds further regularization in the bottleneck. The image is drawn by PlotNeuralNet V1.0.027.
