Given an image pair, TrGAN is able to project them onto the learned transformation space and further extract the semantic variation between them to transform new images.

Unsupervised Image Transformation Learning via Generative Adversarial Networks
Kaiwen Zha, Yujun Shen, Bolei Zhou
arXiv preprint arXiv: 2103.07751

[Paper] [Project Page] [Demo]

In this repository, we propose an unsupervised learning framework, termed as TrGAN, to project images onto a transformation space that is shared by the generator and the discriminator. Any two points in this projected space define a transformation that can guide the image generation process. TrGAN is able to adequately extract the semantic variation between a given image pair and further apply the extracted semantic to facilitating image editing. Some results are shown below.

Changing Season
Source	Target	Sample 1	Sample 2	Sample 3

Adding Clouds
Source	Target	Sample 1	Sample 2	Sample 3

Altering Shape
Source	Target	Sample 1	Sample 2	Sample 3

@article{zha2021unsupervised,
  title   = {Unsupervised Image Transformation Learning via Generative Adversarial Networks},
  author  = {Zha, Kaiwen and Shen, Yujun and Zhou, Bolei},
  journal = {arXiv preprint arXiv:2103.07751},
  year    = {2021}
}