Getting Selected
I’m excited to announce that I’ve been selected for Google Summer of Code 2026 with NumFOCUS, working on the DeepForest project!
My project, “Recovering Historical Image Data Using Automated OrthoRegistration,” focuses on building an automated pipeline to align historical forest health survey annotations with modern satellite imagery.
The Problem
The U.S. Forest Service’s Aerial Detection Survey (ADS) program has produced tens of thousands of hand-drawn polygons marking forest damage from aircraft. Due to GPS imprecision and flight geometry, these polygons are spatially offset by 50-500 meters from their true locations.
This means we can’t directly use these annotations as training data for modern detection models — the labels don’t line up with what’s actually visible in satellite imagery.
My Approach
I’m building a hybrid alignment pipeline that combines:
- Geometric transformations — affine and projective corrections based on known control points
- Learned displacement estimation — a self-supervised deep learning model that predicts per-polygon correction vectors
- Energy-function optimization — combining spectral similarity, edge alignment, and spatial coherence to evaluate registration quality
What’s Next
Over the coming weeks, I’ll be:
- Setting up the development environment and understanding the existing DeepForest codebase
- Prototyping the initial alignment pipeline
- Starting work on the learned refinement layer
Stay tuned for updates!
This is part of my GSoC 2026 blog series. Follow along with the #gsoc tag.