---

name: "dataset-config" description: "Routes NanoDet dataset, preprocessing, and config-validation workflows for COCO, XML, YOLO, and model-assembly setups." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

dataset-config

Use this sub-skill when you need to understand or validate a NanoDet config, prepare a dataset layout, or check the preprocessing path before training or inference.

Use this route for

• Reading or editing YAML config files.
• Verifying save_dir, model, data, device, schedule, evaluator, log, and class_names fields.
• Choosing between CocoDataset, XMLDataset, and YoloDataset.
• Checking keep_ratio, multi_scale, and augmentation pipeline settings.
• Confirming that a config can build the expected model family.

Do not use this route for

• Training / validation / checkpoint lifecycle logic. Use training instead.
• Image/video/webcam inference or model export. Use inference-export instead.
• Repository maintenance or packaging chores.

Read first

• references/configuration.md for the NanoDet config structure and field meanings.
• references/data-formats.md for COCO, XML, and YOLO data-layout details.
• references/troubleshooting.md for config, dataset, and preprocessing failures.
• ../../references/api-reference.md when you need verified builder signatures.

Skill-owned scripts

• scripts/check_config.py — load a config, check class counts, and build the model as a smoke test.
• scripts/check_dataset.py — build a dataset from a config and inspect a sample.

Typical workflow

1. Load the YAML config.
2. Check the model/dataset fields that affect runtime behavior.
3. Verify the dataset layout and annotation format.
4. Run the skill-owned smoke script before moving to training or inference.

Cross-links

• If the config is valid but training fails later, switch to training.
• If the config is valid and you only need to run a demo or export a model, switch to inference-export.