SR Agro Vision
- Python
- PyTorch
- FastAPI
- React
A super-resolution application for agricultural satellite imagery. It trains and compares deep learning models (ESPCN, SwinIR and GAN) on 10 m Sentinel-2 images and evaluates the results with agricultural indices such as NDVI. It includes a command-line pipeline and a web application with an API and maps.
It was the most challenging project. The idea came from NVIDIA's DLSS, in its first versions, which used AI to reconstruct low-resolution images in video games. I asked myself how to take that to something that wasn't a game, and I landed on agriculture: letting a farmer enhance free Sentinel-2 images with neural networks instead of paying for commercial high-resolution imagery.
