SULAND v2 Benchmark: Fixing Annotations Boosts Landmine Detection mAP by ~20 Points
Sagar Lekhak · hf · 2026-08-03
Researchers introduced SULAND v2, a refined RGB dataset and benchmark for surface-landmine detection using UAVs/UGVs. The original dataset suffered from missing annotations, localization errors, and inverted out-of-distribution (OOD) class IDs.
The updated version manually revises labels while preserving original images and splits, comprising 33,771 images and 12,433 bounding boxes. The team benchmarked 35 detector configurations across 9 architectures. Merely fixing annotations boosted YOLOv8 in-distribution (IID) mAP@50 by 14.6-19.6 percentage points and OOD mAP@50 by 25 percentage points.
Results demonstrate that high IID accuracy does not guarantee operational readiness. YOLOv12-Small achieved the highest IID mAP@50 (0.908), while RF-DETR-Large yielded the strongest OOD performance (0.799 mAP@50).
More from Research
- Comma.ai Recaps Hilarious Early Bugs in Self-Driving Models — dosco · 2026-08-03
- Apple-π Dataset Released: Benchmarking Video Models' Physical Reasoning — ziqi_huang_ · 2026-08-03
- Qdrant Open-Sources E-commerce Demo: Millisecond Search Across 5.8M Products — qdrant_engine · 2026-08-03
- Japanese Junior Researchers Launch LENS Community for Computer Vision — HirokatuKataoka · 2026-08-03
- LILO Framework Enables LLMs to Automatically Refactor Code and Generate Docs — burny_tech · 2026-08-03
- NeurIPS 2026 Announces AIDaR Workshop on Data Infrastructure for Scientific AI — arjunrajlab · 2026-08-03