Supporting assets for improving small object detection through resolution, model scale, and NMS threshold tuning.
Reads YOLO label files and computes object sizes in pixels, including the fraction falling below 32px and 16px.
AnalysisReusable training function used for every resolution and model-scale experiment in this article.
ConfigurationCalls comparing the larger model scale and the augmentation settings used for the final experiment.
ConfigurationReference notes on how the NMS IoU threshold affects recall for dense small object scenes.
EvaluationOrdered checklist for diagnosing and improving small object detection performance.
ChecklistGuidance on weighing resolution, model scale, and augmentation changes against inference cost.
GuideSelected resources are designed to support experimentation and practical understanding through focused implementations, utilities, and companion materials accompanying each article.