Berremangra Orange Dataset

The Berremangra orange dataset is a compact, occlusion-intensive image dataset collected from a real Valencia orange orchard at Berremangra Cobbler Farm. It is designed to support research on orange detection and instance segmentation under challenging field conditions, where fruit targets are frequently affected by partial occlusion from leaves, branches, neighbouring fruit, shadows, and cluttered orchard backgrounds. Images were resized to 640 x 640 pixels and annotated in YOLO polygon segmentation format for orange instances. To increase visual diversity, the dataset includes augmented samples generated using random rotations between -15 and +15 degrees and brightness adjustments between -15% and +15%. This mini dataset is intended for evaluating and developing occlusion-aware perception methods, mask refinement strategies, and robust agricultural robotic vision pipelines.