Small Objects, Big Images: Evaluating YOLO and SAHI on 4K Drone Imagery
Standard YOLO inference silently loses small objects when high-resolution drone images are downscaled to the model's input size. This post walks through how the AgriDrone Vision pipeline uses SAHI sliced inference and a COCO-based evaluation layer to make that trade-off measurable — and why the real deliverable is a georeferenced map, not a bounding box.
Read More →