GAN, generate regression output by the real image, not from the random noise

Question

Is this concept possible to be implemented with the GAN algorithm?

I want the GAN to generate a regression-output(G-Value) of the shape(4,) by the real-image, not from th