Project Overview
- Period: 2023.07 ~ 2023.09
- Organization: 국방기술진흥연구소
- Role: Research Intern at SI Analytics (SR Team)
- Team Size: 4
Main Responsibilities
- Research on Blind Super Resolution
- Generating synthetic Low resolution for training in paired (hr, lr) datasets
- Use various SR models to verify enhancement of model performance
- Research on estimation of degradation in low resolution images (SAR & EO)
- Use Frechet distance to estimate synthesize data and real lr data
- Using various kinds of kernels (Noise, Blur, etc.) to estimate speckle noise in SAR datasets
- Implement from scratch to whole pipeline to learn degradation and generate in end-to-end manners
Technical Skills
- Python (osgeo, gdal, tifff)
- PyTorch
- TensorFlow
- QGIS
Project Description
- Develop a method for super-resolution using only real low-resolution and high-resolution images in an unpaired dataset
- Improved existing super-resolution (SR) models by estimating unknown degradations present in Synthetic Aperture Radar (SAR) or Electro-Optical (EO) imagery
- Implemented and experimented with a paper focused on improving super-resolution (SR) model performance