-
Detecting Hail Damage on Rooftops
Building a dataset and training YOLO models to automatically detect hail damage on roof shingles from drone imagery.
-
From Point Regression to Roof Slope Detection
Training vision transformers to predict roof slope locations using point regression, Hungarian matching, and custom loss functions.
-
Counting Roof Faces: A Dataset of 890K Buildings
How we built a large-scale dataset of 890K annotated buildings for roof slope detection from drone and satellite imagery.
-
Map-free Visual Relocalization
Exploring MASt3R, 4M, and depth-augmented approaches for the Map-free Visual Relocalization Challenge 2024.
-
Shingle Damage Detection: From V1 to V2
Iterating on a hail and shingle damage detection model to handle multiple roofing materials, reduce false positives, and work across diverse conditions.