超越聚合分数：GWHD 小麦穗检测模型库的按国家领域偏移分析
原标题：Beyond the Aggregate Score: Per-Country Domain Shift in the GWHD Wheat Head Detection Model Zoo
AI 摘要
Hugging Face 博客发布了对 9 个小麦穗检测模型（YOLOv8、YOLOv11、YOLOv26、RF-DETR）在 GWHD 2021 数据集上的按国家分层评估结果。所有模型在中国子集上表现最佳，而多数 YOLO 模型在美国子集上表现最差，RF-DETR 则在澳大利亚表现最差，表明聚合分数掩盖了显著的领域偏移。该分析基于自建的 per-image 元数据，揭示了模型在不同地理区域的性能差异，对实际部署具有重要参考价值。
正文节选
A few weeks ago I open-sourced 9 object detection models — YOLOv8, YOLOv11, YOLOv26, and RF-DETR, spanning nano through x-large variants — fine-tuned on the Global Wheat Head Dataset (GWHD) 2021, a dense, single-class wheat-head detection benchmark assembled from field images captured across 6 countries and 18 research institutions to maximize genotype, growth-stage, and imaging-condition diversity. All 9 models were trained and evaluated under one shared pipeline (DetectionBench), with the usua