CycleGAN is a type of generative adversarial network that enables image-to-image translation between two domains without requiring paired examples.
This project applies CycleGAN to translate images between horses and zebras, learning to convert one to the other while preserving key features like shape and texture.
The model utilizes two generators and two discriminators to enforce a cycle consistency loss, ensuring that translated images can be converted back to their original form.