Theses and Dissertations

Advisor

Chen, Jingdao

Committee Member

Gudla, Charan

Committee Member

Chen, Zhiqian

Date of Degree

8-7-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Graduate Thesis - Open Access

Major

Computer Science (Artificial Intelligence)

Degree Name

Master of Science (M.S.)

College

James Worth Bagley College of Engineering

Department

Department of Computer Science and Engineering

Abstract

Restoring severely blurred images remains a significant challenge in computer vision, impacting applications in autonomous driving, medical imaging, and photography. This thesis introduces a novel training strategy based on curriculum learning to improve the robustness of deep learning models for extreme image deblurring. Unlike conventional approaches that train on only low to moderate blur levels, this method progressively increases the difficulty by introducing images with higher blur severity over time, allowing the model to adapt incrementally. Additionally, perceptual loss and hinge loss were integrated during training to enhance fine detail restoration and improve training stability. Various curriculum learning strategies were experimented with, and the impact of the train-test domain gap on the deblurring performance was explored. This architecture-agnostic framework can be applied to various generative models, including GAN-based and Transformer-based networks. Experimental results on the Extreme-GoPro and Extreme-KITTI datasets demonstrate that this approach significantly outperforms existing state-of-the-art methods in handling extreme blur scenarios while maintaining high visual fidelity. Both qualitative and quantitative analyses confirm that curriculum learning substantially improves deblurring performance, making it a promising direction for future research on mitigating image artifacts.

Sponsorship (Optional)

Intelligence Advanced Research Projects Activity (IARPA) via Department of Interior/ Interior Business Center (DOI/IBC) contract number 140D0423C0075

Download

Share

COinS