Image-to-Image translation is the task of translating images between domains while maintaining the identity of the images. The task can be used for entertainment purposes and applications, data augmentation, semantic image segmentation, and more. Generative Adversarial Networks (GANs), and in particular Conditional GANs have recently shown incredible success in image-to-image translation and semantic manipulation. However, such methods require paired data, meaning that an image must have ground-truth translations across domains. Cycle-consistent GANs solve this problem by using unpaired data. Such methods work well for translations that involve color and texture changes but fail when shape changes are required. This research analyzes the trade-offs between the cycle-consistency importance and the necessary shape changes required for natural looking imagery. The research proposes simple architectural and loss changes to maintain cycle-consistency strength while allowing the model to perform shape changes as required. The results demonstrate improved translations between domains that require shape changes while preserving performance between domains that don’t require shape changes.
Library of Congress Subject Headings
Image processing--Digital techniques; Neural networks (Computer science); Imaging systems--Image quality
Computer Engineering (MS)
Department, Program, or Center
Computer Engineering (KGCOE)
Longman, Ram, "Embedded CycleGAN for Shape-Agnostic Image-to-Image Translation" (2018). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus