Eco-Restoration

Eco-Restoration is a computer-vision pipeline for analysing ecosystem-restoration imagery: detecting species, segmenting damaged areas, and using generative inpainting to visualise what a restoration intervention would look like at a specific site.

The detection stage runs YOLOv8 on drone and satellite imagery. The segmentation stage runs the Segment Anything Model (SAM) on the detected regions to produce pixel-accurate masks. The generative stage runs Stable Diffusion in an inpainting configuration conditioned on the segmentation mask, so the model “fills in” only the damaged area rather than regenerating the whole image.

The whole pipeline is wrapped in a FastAPI service that exposes a single endpoint: upload an image, get back a JSON with detected species plus a base64-encoded “restored” version of the image. The model code, the FastAPI wrapper, and a small React dashboard that calls the API ship in the project.

This entry does not link to a public repository. The work was done on a private codebase and the case study here is the technical shape, not a live demo.