Freeform optical design with a generative AI model based on a conditional generative adversarial network

In this paper, we proposed an artificial intelligence (AI)-based optical design framework of freeform imaging systems using a conditional generative adversarial network, which has outstanding applications in fields such as image generation. The adversarial and competitive training process of both the generator and the discriminator networks for optical design tasks, with the data-driven loss and physics-informed loss of the optical design model, is proposed and used. The generator network after training can be taken as a generative AI model of highly authentic freeform imaging systems. For specific design tasks, after the system parameters and folding geometry parameters (conditional information) and random noise vectors are input into the generator, freeform systems fulfilling the design requirements with various structures can be generated immediately. These systems can be taken as very good starting points for further rapid optimization. The design efficiency of the freeform optical design is dramatically improved, and the dependence on human design experiences and existing references is reduced. Several examples are presented to show the feasibility of the proposed framework. The work proposed in this paper significantly advances the development of optical design with generative AI.