Pixelating to the Edge - Generative AI Art on Edge Devices

Pardesi, Sonal Deepak (2024) Pixelating to the Edge - Generative AI Art on Edge Devices. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (12MB) | Preview
Preview	PDF (Configuration Manual) Download (4MB) | Preview

Abstract

Generative AI is now changing the landscape of how people are getting things done. In this study we are focusing on one of the most popular application of generative AI which is text to image generation, a technology which created huge craze in 2021. Even after 3 years, it is not as effective and speedy on the edge devices like mobile as its own web version. In this research, we will compare a number of model variations and analysis what factors affect the inference speed of the image generation. Currently Mobile diffusion, although commercially not available, has claimed that it can achieve 0.02 seconds of speed. I tried to study if any architectural changes and sampling techniques can improve the inference speed while maintaining quality of image. The changes in the sampling operations like changing the scheduler from PNDMS to DDIM gave a 6.57 percent increase in inference speed with a bit of degradation in FID and CLIP score. The architectural changes gave significant improvement of up to 15.24 percent increase and good results with FID and CLIP score. This research will explore the much needed generative AI technology’s requirement on edge devices.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Simiscuka, Anderson UNSPECIFIED
Uncontrolled Keywords:	Generative AI; GANs; text-to-image
Subjects:	N Fine Arts > NX Arts in general Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence
Divisions:	School of Computing > Master of Science in Artificial Intelligence for Business
Depositing User:	Ciara O'Brien
Date Deposited:	02 Jul 2025 17:17
Last Modified:	02 Jul 2025 17:17
URI:	https://norma.ncirl.ie/id/eprint/7996

Actions (login required)

View Item