Researchers have released SULAND_v2, an improved dataset for training AI models to detect surface landmines from RGB images captured by unmanned aerial and ground vehicles. The original SULAND dataset contained numerous annotation errors, including missing or false labels, localization inaccuracies, and inconsistent class identification. SULAND_v2 corrects these issues, providing 33,771 images with 12,433 bounding boxes for detecting specific mine types like PFM-1 and PMA-2. Benchmarking various object detection models on this refined dataset showed significant improvements in accuracy, with YOLOv12-Small achieving the highest in-distribution performance and RF-DETR-Large excelling in out-of-distribution scenarios, highlighting the importance of robust generalization for operational readiness. AI
IMPACT Enhances the reliability of AI systems for critical landmine detection tasks, improving operational readiness and safety.
RANK_REASON Publication of a refined dataset and benchmark for a specific AI task (object detection for landmine detection). [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
