Recipe 01 — Bootstrap A Dataset

1. Run dataset_inspect to confirm image count, class count, and split structure.
2. Run dataset_validate to catch unreadable images, bbox range issues, and class-id errors.
3. Run dataset_find_duplicates if the source is stitched from multiple exports.
4. If needed, run dataset_convert or dataset_split before training.

This recipe is the fastest way to turn “I got a folder from someone else” into a trusted local training input.