Rect2Sheet: A Dataset of Sheet Metal Connection Designs
收藏资源简介:
Description A synthetic human-validated dataset of 19,231 sheet-metal designs pairing rectangle layouts with their sheet-metal solutions, plus a parallel set of discarded examples. Additionally provided are also datasets with 80,028 unvalidated designs. Test datasets can be found in the project's GitHub: https://github.com/atsmt/Rect2Sheet Built for training/evaluating models that turn tabs into bent parts. Generated with the SheetGen repository: https://github.com/chriswitt77/SheetGen Contents Folder Contents dataset_json Human-validated JSON dataset dataset_discarded_json Rejected (unmanufacturable) JSON dataset dataset_fs Same dataset in FeatureScript dataset_discarded_fs Rejected dataset in FeatureScript dataset_80k_json Bigger but unvalidated dataset dataset_80k_discarded_json dataset_80k_fs dataset_80k_discarded_fs Each design is a zero-padded triple: 00001_rectangle.json # input: rectangle tabs (corner points A/B/C, optional mounts) 00001_solution.json # target: fold sequence, bends (tabs/points/direction), folded tabs 00001_metadata.json # id, rectangle_id, num_rectangles, has_mounts, allow_non_90, has_split_tabs How it was generated Synthetic pipeline in `src/dataset`, configured via `config/config.yaml`: Rectangles: Random connected layouts of 2–4 tabs with varied sizes, gaps, angles (incl. non-90°), and optional mount holes. Solutions: The SheetGen engine explores fold topologies and produces candidates; manufacturability filters (tool collision, unfolding overlap, thin segments) split them into **accepted** vs **discarded**; up to 12 diverse solutions per design are kept. For solutions with mounts, input rectangles can also be split into two. Verification: The author manually inspected all generated solutions and discarded edge cases (for example designs with too thing parts). Packaging: Pairs accepted solutions with their rectangles, strips simulation-only fields, and emits clean JSON triples + FeatureScript. Credits Generated with SheetGen https://github.com/chriswitt77/SheetGen. Please cite the Rect2Sheet dataset and SheetGen repository when used.



