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。
data-configuration【数据与研究】Configure MMSegmentation configs, datasets, transforms, class palettes, and safe dataset layout checks.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Use this sub-skill when an MMSegmentation task is about experiment config files, MMEngine config inspection, dataset layout preparation, dataset classes, transform pipelines, class metadata, palettes, or dataset conversion planning.
_base_ inheritance, checking config naming, or applying --cfg-options overrides before a run.data_prefix, ann_file, and BaseSegDataset arguments.LoadAnnotations, augmentation transforms, PackSegInputs, class subsets, palettes, ignore_index, or reduce_zero_label.model-index.yml, dataset-index.yml, and native dataset converter patterns without depending on source checkout scripts.references/configuration.md for config inheritance, naming, inspection, and override rules.references/datasets-and-transforms.md for dataset layout, BaseSegDataset, transform, class, palette, and label-index contracts.references/dataset-conversion.md before converting raw datasets; converters are dataset-specific and reference-only here.scripts/inspect_mmseg_config.py --help and scripts/check_dataset_layout.py --help to inspect configs and validate dataset file pairing safely.references/troubleshooting.md before editing configs or data files.scripts/inspect_mmseg_config.py expands an MMEngine config, applies optional --cfg-options, prints selected keys, and only writes output when --dump is explicitly provided.scripts/check_dataset_layout.py checks image/annotation directory existence, suffix counts, expected image-to-mask pairing, optional split files, and label-index notes without importing MMSegmentation.