| MS Seminar


Name of the Speaker: Mr. AMAN KUMAR (EE22S037)
Guide: Dr. Rajagopalan A N
Venue: ESB-244 (Seminar Hall)
Online meeting link: http://meet.google.com/oif-owyf-kum
Date/Time: 2nd July 2025 (Wednesday), 10:30 AM
Title: Self-Supervised Image Deraining and Realistic Rain Synthesis Using Visual Autoregressive Models.

Abstract :

Rain significantly degrades the visual quality of outdoor images, affecting the performance of various vision-based downstream tasks such as object detection, tracking, and scene segmentation. Most existing single-image deraining methods rely on supervised learning with paired rainy-clean images, which are difficult to collect due to the dynamic nature of rain and environmental changes. Consequently, many models are trained using synthetic rain data, leading to domain gaps that compromise the generalization ability of trained models in real-world settings. To address these challenges, we propose two solutions: (1) A self-supervised learning framework for image deraining using monocular real-world video sequences. (2) A Visual AutoRegressive (VAR) generative model for synthesizing realistic rain on clean images to construct more effective training datasets.