Image Super-Resolution with ESRGAN
Upscaled medical/satellite imagery while preserving fine details.
Problem
Low-res scans limited diagnostic and analytic utility.
Approach
- Trained ESRGAN for x2/x4 upscaling with curated augmentations
- Balanced perceptual/adv losses to preserve edges and texture
Impact
- Improved clarity with a 30% PSNR boost on validation imagery
- Enabled better readability of fine structures in scans/maps