-
Building a Multi-Task Change Detection Dataset
How we built QFabric, a multi-temporal satellite dataset for tracking urban construction changes across 100 cities.
-
Swimming Pool Identification from Satellite Imagery
Benchmarking YOLOv3 and SAM-GEO for detecting residential swimming pools in high-resolution satellite images over Belo Horizonte, Brazil.
-
Detecting Trees in Urban Spaces from Aerial Imagery
Comparing YOLO V3, YOLO V5, and SAM-GEO for automated tree detection in high-resolution UAV imagery.
-
How Well Do Zero-Shot Models Work on Satellite Data?
An extensive benchmark of CLIP, BLIP, and OpenCLIP foundation models on satellite and aerial image classification tasks, revealing surprising gaps and prompt engineering strategies.
-
Detecting Overhanging Trees on Residential Roofs
Exploring multiple approaches---from UNet segmentation to unsupervised methods---for identifying trees that overhang roof structures in aerial imagery.