AI@UGA | Spring 2026
End-to-end computer vision pipeline using YOLOv8 and Mask R-CNN to detect, count, and estimate volume of hurricane-damaged fallen trees from 7,357 orthorectified drone images (62.7 GB).
Collaboration with Warnell School of Forestry and Natural Resources, University of Georgia.
fallen-tree-detection/ ├── data/ │ ├── raw/ # DO NOT COMMIT │ ├── samples/ # sample images for testing │ └── annotated/ # exported from Roboflow ├── src/ │ ├── detection/ # predict.py │ ├── measurement/ # volume.py │ ├── registration/ # dedup.py │ └── pipeline/ # dataloader.py, reporter.py ├── notebooks/ # Jupyter EDA notebooks ├── docs/ ├── run_pipeline.py └── requirements.txt
pip install -r requirements.txt
python run_pipeline.py --input data/samples/ --output results/
7,357 orthorectified GeoTIFF images across 17 flight zones. Captured April 26, 2025, at New York Road, Georgia. Raw data stored separately - not in this repo.