Home / Current Issue / Paper 1711962
Low-Power Real-Time Image Enhancement on Edge Devices: A Study of Lightweight GANs and Quantization Effects
Subject area: Science,Engineering and Technology · Area of research: Edge AI and Image Processing
DOI: https://doi.org/10.64388/IREV6I8-1711962
Abstract
This paper presents an in-depth study on low-power, real-time image enhancement techniques designed specifically for edge devices. By leveraging lightweight Generative Adversarial Networks (GANs) and model quantization, we explore methods to achieve high-quality visual enhancement under strict computational and memory constraints. The research evaluates multiple GAN architectures and quantization strategies in terms of power efficiency, latency, and perceptual image quality, providing a comprehensive comparison across embedded AI platforms. Our analysis extends beyond traditional benchmarking to include real-world constraints such as voltage-frequency scaling, thermal throttling, and on-device inference limitations.
How to cite this paper
@article{1711962,
author = {Adit Shah, Mathew Campisi, Arpit Arora},
title = {Low-Power Real-Time Image Enhancement on Edge Devices: A Study of Lightweight GANs and Quantization Effects},
journal = {Iconic Research And Engineering Journals},
year = {2023},
volume = {6},
number = {8},
pages = {399-404},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1711962.pdf},
abstract = {This paper presents an in-depth study on low-power, real-time image enhancement techniques designed specifically for edge devices. By leveraging lightweight Generative Adversarial Networks (GANs) and model quantization, we explore methods to achieve high-quality visual enhancement under strict computational and memory constraints. The research evaluates multiple GAN architectures and quantization strategies in terms of power efficiency, latency, and perceptual image quality, providing a comprehensive comparison across embedded AI platforms. Our analysis extends beyond traditional benchmarking to include real-world constraints such as voltage-frequency scaling, thermal throttling, and on-device inference limitations.},
month = {February},
doi = {https://doi.org/10.64388/IREV6I8-1711962}
}