VGER-LR Strawberry Occlusion Dataset and Code Snapshot

This upload provides the dataset and code snapshot supporting the study on visibility-guided localization-robust amodal detection for densely occluded strawberries. The dataset contains greenhouse strawberry images with YOLO-format amodal bounding-box annotations for ripe and raw strawberries. The bounding boxes represent full-object/amodal boxes rather than visible-region boxes. Occlusion-level annotations and visibility metadata are included to support occlusion-stratified evaluation and visibility-supervised training. The training split contains both natural images and offline-generated synthetic occlusion images, while the validation split contains natural greenhouse images. The code snapshot includes the VGER-LR implementation, configuration files, trained weights, data generation scripts, evaluation scripts, metric files, and documentation needed to reproduce the main experiments.