Processor-implemented methods and systems for aerodynamically optimizing a design geometry of a vehicle body using a convolutional neural network (CNN) are provided. The method may include receiving a signed distance function (SDF) data file that represents the design geometry of the vehicle body. The method includes receiving a range of inflow boundary conditions. The processor processes the SDF over the range of boundary conditions, using the CNN, to generate therefrom drag and lift outputs for the design geometry. The drag and lift outputs may be displayed in the form of one or more intensity maps.