Researchers have introduced SULAND_v2, a refined RGB dataset and benchmarking framework designed to improve surface landmine detection (specifically PFM-1 and PMA-2) for UAV and UGV platforms under domain shift. The original SULAND dataset suffered from significant technical flaws, including localization errors, inconsistent visibility criteria, and an inverted out-of-distribution (OOD) class-ID convention. To resolve these, the authors manually revised annotations to ensure labeling validity and spatial precision, resulting in a comprehensive dataset comprising 33,771 images and 12,433 bounding boxes. The study benchmarked 35 detector configurations across nine different model families. The impact of these refinements was substantial: fixing the OOD class-ID convention boosted the mean YOLOv8 OOD mAP@50 by approximately 25 percentage points, while improved annotations increased the YOLOv8 in-distribution (IID) mAP@50 by 14.6–19.6 percentage points. Performance analysis revealed a discrepancy between IID and OOD capabilities; YOLOv12-Small achieved the highest IID mAP@50 at 0.908, whereas RF-DETR-Large demonstrated superior robustness in OOD scenarios with a 0.799 mAP@50 and a 0.675 recall. This research highlights that high in-distribution accuracy does not necessarily translate to operational readiness in varying deployment conditions, establishing SULAND_v2 as a vital benchmark for evaluating domain-shift robustness in mine-action survey support.