Accuracy of the convolutional ResNet autoencoder described in Sect. 4.2. Left: (x, y) predicted coordinates of the BCGs as a function of their true (x, y) coordinates as given in the cosmoDC2 simulation. The figure shown here is obtained for the BCGs in the evaluation sample. The colour bar represents the number density of points, with yellow representing higher density. Right: Histogram of the difference between the predicted and true (x, y) coordinates of the BCGs in the evaluation sample as detected by the ResNet autoencoder (top). The red and blue histograms represent the x and y coordinates respectively. Accuracy of the ResNet autoencoder depending on the defined resolution considered for a well-detected BCG (bottom). The blue and red lines correspond to an accuracy of 80% and 90%, respectively.