Visualization of the baseline and residual architecture for the CNN lensfinder: the convolution blocks (red) indicate thesize of the kernel and the number of features. The fully connected blocks (yellow) indicate the number of features. The arrows indicate the flow of the data, and between the blocks, we show the dimensionality of the input (N_pixel × N_pixel × N_features). The last fully connected layer yields a confidence value of the object being a lens. The initial layer has N_b features, either one or four, depending on the category of the data (space and ground, respectively). Batch normalization and dropoutlayers are indicated as gray blocs.