i train a generative network to produce pictures from a few input variables.
right now it works suprisingly well. All generated Pictures have a shape which are not distingquishable to original.
My Problem is, that the Position of the Object in the Picture is more important to me than the shape. but every generated Object is is located in or close to the center.
Do you have any idea how to train a Network, which is more sensitive to the Position of the Object?