# Resolution Dependant GAN Interpolation for Controllable Image Synthesis Between Domains

**Authors:** Justin N. M. Pinkney, Doron Adler  
**arXiv:** [2010.05334](https://arxiv.org/abs/2010.05334)  
**Date:** 2020-10-11  
**Summary page:** https://www.summarizepaper.com/en/paper/2010.05334v1/

## Key points

- Limitations of Generative Adversarial Networks (GANs) in generating images from novel domains
- Proposal of a resolution-dependent interpolation method for generative models based on the StyleGAN architecture
- Ability to generate images from entirely new domains while providing control over the output
- Potential applications in creative fields where generating imagery from unique domains is crucial
- Use of the proposed method for controllable image synthesis, enabling artists and designers to explore new visual possibilities
- Experimental results and figures demonstrating the effectiveness of the technique
- Advancement of GAN capabilities by enabling them to generate images from novel domains with control over the output
- Promising opportunities for creative applications and expanded potential use cases for GAN technology

## Summary

The paper titled "Resolution Dependant GAN Interpolation for Controllable Image Synthesis Between Domains" discusses the limitations of Generative Adversarial Networks (GANs) in generating images from novel domains and proposes a method to overcome this challenge. While GANs are capable of generating photo-realistic images within their training data domain, they lack the ability to generate imagery from completely new domains. Additionally, users often desire a level of control and artistic direction in the generated output rather than random results. To address these issues, the authors present a resolution-dependent interpolation method for generative models based on the StyleGAN architecture. This approach allows for the generation of images from entirely novel domains while providing control over the nature of the output. By interpolating between different generative models, users can achieve a desired level of creativity and artistic direction in their synthesized images. The proposed method has potential applications in various creative fields where generating imagery from unique domains is crucial. The authors provide insights into how their approach can be used for controllable image synthesis, enabling artists and designers to explore new visual possibilities. The paper also includes experimental results and figures to demonstrate the effectiveness of their technique. In conclusion, this research contributes to advancing the capabilities of GANs by enabling them to generate images from novel domains with a degree of control over the output. The resolution-dependent interpolation method presented in this paper offers promising opportunities for creative applications and expands the potential use cases for GAN technology.