Method for training a neural network

A method for training a neural network for image transformation, comprising:

• providing a vehicle on which are mounted a first camera and a second camera, the first camera and the second camera each having a same orientation relative to the vehicle and being arranged spaced apart by a distance d in a default moving direction of the vehicle, wherein the first camera and second camera are different from each other with respect to at least one property among the group of properties consisting of maximum pixel resolution, color gamut, use of color or polarization filters, camera manufacturer, camera model, and exposure settings;
• allowing or controlling the vehicle to move along a trajectory;
• while moving along the trajectory, repeatedly performing a recording operation for obtaining a dataset comprising a plurality of pairs of images, wherein each recording operation comprises triggering the first camera and the second camera at different time instants such that at a time of recording an image, the entrance pupil of the first camera and the entrance pupil of the second camera are at substantially the same position along the trajectory, wherein each pair of images comprises an image (Im1) recorded by the first camera and an image (Im2) recorded by the second camera during a given recording operation;
• wherein for the purpose of recording an image by the first camera and the second camera such that the entrance pupils of those cameras are at substantially the same position at the time of recording the respective images, triggering of a camera among the first camera and second camera of which the entrance pupil reaches said same position the later is delayed by an amount that is equal to the distance d between the first camera and second camera in the default moving direction divided by the velocity v of the vehicle relative to the triggering of another camera among the first camera and second camera;
• and
• using the obtained dataset for training a neural network to learn an image transformation for transforming an image of a scene recorded by the first camera into an image that mimics an image of the scene as it would have been recorded by the second camera.