Autoencoder reconstructs shape but signal amplitude it's way off

Question

I\'m having an issue with a convolutional autoencoder, essentially it has to learn how to reconstruct the input 2D array, the problem is that the general shape of the array