Code and data for "Recognition of grazing behavior in complex environments based on the YOLO model"

<p dir="ltr">Dataset, code, and trained models for the manuscript "Recognition of grazing behavior in complex environments based on the YOLO model".</p><p><br></p><p dir="ltr">Contents:</p><p dir="ltr">(1) zenodo_package.rar — dataset v2 (1,200 train / 200 validation / 200 sequence-held-out test images, four behavior classes: grazing, walking, resting, human interference; YOLO format with annotation provenance); pose keypoint datasets in silver (model-assisted) and fully human-verified (gold) versions, six keypoints per sheep; trained YOLOv8x detector and YOLOv8n-pose weights with evaluation artifacts; dataset and pose-track reports.</p><p dir="ltr">(2) zenodo_code.rar — the complete experiment pipeline and reproduction scripts (dataset build, training matrix, ablation, pose track, BoT-SORT tracking, cascade evaluation), with CPU/GPU requirements and a reproduction guide.</p><p><br></p><p dir="ltr">All reported results regenerate from these scripts plus the deposited data (fixed random seed 0).

Related video samples are hosted at doi.org/10.6084/m9.figshare.32824283.</p><p><br></p>