Figure 3

Architecture of the applied autoencoder. Numbers describe the shapes of computed feature maps. Convolutional layers ('conv’) are comprised of convolutional filters (light orange) and Leaky ReLU (\(\alpha =0.01\)) activation functions (orange). Spatial downsampling is performed using max-pooling layers (red), resulting in a set of bottleneck features. Upsampling operations ('up’, blue), and convolutional layers are then used to reconstruct the input image. A sigmoid activation (purple) is used as model output to match the range of the input data.