Research

Recovering Historical Image Data Using Automated OrthoRegistration

Google Summer of Code 2026 · NumFOCUS / DeepForest

May 2026 — Present
  • Building an automated pipeline to align U.S. Forest Service Aerial Detection Survey (ADS) polygons with NAIP satellite imagery using deep learning
  • Developing a learned refinement layer using self-supervised displacement estimation to correct systematic spatial offsets of 50-500 meters
  • Scaling the pipeline to process ~48,000 Oregon ADS polygons to generate weak training labels for forest health detection models
  • Contributing to the open-source DeepForest Python package used by researchers worldwide

DeepForest vs DeepTrees — Comparative Analysis

Independent Research · University of Education

Apr 2026
  • Conducted a technical benchmarking study comparing DeepForest and DeepTrees for tree crown detection
  • Analyzed architectural differences, training strategies, and prediction accuracy across multiple datasets
  • Produced a structured Jupyter Notebook documenting the methodology and findings

Education

2025
MS Information Technology University of Education, Lahore
2021
BS Information Technology University of Education, Lahore