技能创建器
skill-creator
创建、改进和评估自定义技能,把专业流程转化为可重复使用的 AI 能力。
segmentation【数据与研究】Use this YOLOv5 sub-skill for instance-segmentation training, validation, prediction, mask data, segmentation checkpoints, and mask troubleshooting.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Use this route for YOLOv5 instance segmentation workflows involving segment/predict.py, segment/train.py, segment/val.py, *-seg.pt checkpoints, segmentation YAMLs, polygon labels, masks, or mask-specific failures.
references/workflows.md for segmentation prediction command shapes, --retina-masks, output directories, and source handling. Use scripts/plan_segmentation_command.py before running commands.coco128-seg.yaml, custom segmentation labels, pretrained vs scratch segmentation configs, DDP, and output planning.--overlap, --mask-ratio, confidence/IoU, and exported-format validation handoff.references/data-formats.md before converting detection labels or assuming a dataset is segmentation-ready.yolov5s-seg.pt; do not use plain detection checkpoints for mask workflows unless the task explicitly builds a new segmentation model.models/segment/yolov5*-seg.yaml when training from scratch.coco128-seg.yaml or a tiny local fixture for smoke work.--project and --name explicit because prediction/training/validation write run artifacts.references/datasets-and-weights.md for segmentation checkpoint names and dataset download risks.../detection/SKILL.md and classification to ../classification/SKILL.md.../export/SKILL.md.../serving/SKILL.md only if the service itself is the user's task.