on December 11, 2025 from 4:30 pm to 6:30 pm
Divya Mishra, Assistant Professor at the University of Petroleum and Energy Studies (Dehradun, India), invited by Tanmoy Mondal (GT Image - FOX), will give a seminar.
Title: Real-World Unsupervised Image Super-Resolution
Image super-resolution (SR) enhances images by recovering fine spatial details, but real-world data often suffer from sensor limitaions, variable degradations, and a lack of paired high/low-resolution (HR/LR) training sets. The developed methods overcome key limitations, including the absence of paired training data, insufficient training samples, and limited dataset diversity, while demonstrating strong generalization across various image domains, from satellite imagery to natural real-world images. The research presents work in unsupervised or semi-supervised settings, including Self-FuseNet, which uses multi-modal fusion from a single low-resolution image without prior training; CLSR, which employs contrastive learning to distinguish similar and dissimilar image features; NST-CL, which combines contrastive learning with neural style transfer; and DRSR, which learns degradation representations using internal patch-recurrent patterns. Further, the developed methods are also used in an agricultural application to enhance microscopic root-hair imaging.
Amphitéâtre IRCICA, 50 Avenue Halley, 59650 Villeneuve d'Ascq
français
English