Patch-Based Generative Reconstruction of Task-Oriented Images for Bandwidth-Efficient Communication

Wanninayaka Tennakoon Mudiyanselage, Sameera Sudath Tennakoon (2025) Patch-Based Generative Reconstruction of Task-Oriented Images for Bandwidth-Efficient Communication. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (4MB) | Preview
Preview	PDF (Configuration Manual) Download (798kB) | Preview

Abstract

In bandwidth limited environments such as space and underwater explorations, search and rescue missions, etc., sending full images is challenging. Limited data rates, power and signal issues make it hard to transmit full resolution images. This research addresses the gap in current techniques by presenting a novel, bandwidth usage efficient pipeline that balances data reduction with image quality. Identifying and sending only the important parts of an image can significantly reduce the amount of data required for transmission. These parts are chosen using Saliency maps from task specific models. The selected parts are combined into a stitched grid and compressed by sending only the codebook indices of Vector-Quantized Generative Adversarial Network (VQ-GAN). On the receiver side, VQ-GAN decoder and pretrained Masked Autoencoder (MAE) fill the missing regions. Experimental results show that the reconstructs images with higher visual quality and improved pixel-wise metrics than the baseline. With VQ-GAN compression, images still closely match baseline performance while achieving compression ranging 20× to 160×, demonstrating the effectiveness of this method under severe bandwidth constraints.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Sahni, Vikas UNSPECIFIED
Subjects:	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
Depositing User:	Ciara O'Brien
Date Deposited:	02 Sep 2026 11:07
Last Modified:	02 Sep 2026 11:07
URI:	https://norma.ncirl.ie/id/eprint/9772

Actions (login required)

View Item