Two-Stage Domain Adapted Training For Better Generalization In Real-World Image Restoration And Super-Resolution
Authors: Cansu Korkmaz, A. Murat Tekalp, Zafer Dogan
Venue: IEEE International Conference on Image Processing (ICIP) 2021, Anchorage, AK, USA, September 19-22, 2021
DOI: 10.1109/ICIP42928.2021.9506380
Overview
This work proposes a two-stage domain adaptation training strategy that significantly improves generalization performance for real-world image restoration and super-resolution applications. The method first adapts to synthetic domains and then fine-tunes on real-world data, enabling better transfer of learned representations to practical scenarios.
Key Contributions
- Two-stage domain adaptation framework
- Improved generalization to real-world scenarios
- Better performance on practical image restoration tasks