{"success":true,"data":{"author":"CADGenBench team (HuggingAI4Engineering), CAD geometry from Mecado","bucketAccessType":"public","bucketName":"HuggingAI4Engineering/cadgenbench-data","bucketPrefix":"","category":"Code & Software Engineering","cloudProvider":"huggingface","createdAt":"2026-11-21T22:15:33.222Z","description":"Public task inputs for CADGenBench, a benchmark of how well AI systems produce correct 3D mechanical parts as STEP solids, run by the HuggingAI4Engineering team. It has 81 fixtures of real mechanical parts with mating interfaces such as locating jigs, bolt patterns and slots, sourced from Mecado. 49 generation fixtures ask a system to reproduce a part as an accurate 3D solid from an engineering drawing (one or more PNG images plus a text prompt); 32 editing fixtures give a starting STEP solid, a watertight mesh sidecar (input.mesh.npz), iso, front, top and right renders and a written instruction for a design change. Each fixture is a numbered folder with description.yaml (prompt, task type, input files, modality). Ground truth is held privately; submissions (one output.step per fixture) are scored on a leaderboard Space by a validity gate followed by shape similarity, interface match and topology match. About 360 files, 0.2 GB (measured 2026-11-21); open STEP files with CadQuery, OpenCascade or FreeCAD. This is an evaluation set, not a training corpus. Released under the Open Data Commons Attribution License (ODC-BY); see github.com/huggingface/cadgenbench.","displayTitle":"CADGenBench Inputs: 81 Mechanical-Part Fixtures for AI CAD Generation from Engineering Drawings and STEP Editing (Hugging Face)","huggingFaceUrl":"https://huggingface.co/datasets/HuggingAI4Engineering/cadgenbench-data","id":"eb9d783f-22f3-5889-ad45-8a8d9574b3f0","license":"ODC-By-1.0","licenseEvidenceUrl":"https://huggingface.co/datasets/HuggingAI4Engineering/cadgenbench-data","objectCountEstimate":360,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"HuggingAI4Engineering/cadgenbench-data\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"HuggingAI4Engineering/cadgenbench-data\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=HuggingAI4Engineering, license:odc-by, size_categories:n<1K, downloads=25340"],"format":"mixed","generatedAt":"2026-11-21T22:31:33.804Z","layout":{"example":"201/input.mesh.npz","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"201/input.mesh.npz","truncatedAt":512}],"sampleObjects":["101/description.yaml","101/input.png","102/description.yaml","102/input.png","103/description.yaml"],"sidecars":[{"bytes":"---\nlicense: odc-by\nlanguage:\n- en\npretty_name: CADGenBench (Inputs)\ntask_categories:\n- image-to-3d\n- text-to-3d\ntags:\n- cad\n- 3d\n- step\n- mechanical-engineering\n- benchmark\n- cad-generation\n- cad-editing\nsize_categories:\n- n<1K\nannotations_creators:\n- expert-generated\nlanguage_creators:\n- expert-generated\nmultilinguality:\n- monolingual\nsource_datasets:\n- original\nviewer: false\n---\n\n# CADGenBench (Inputs)\n\nPublic inputs for the **CADGenBench** benchmark, which measures how well AI\nsystems produce correct 3D mechanical parts as STEP files. This repository\nholds the task inputs only; the ground truth is withheld in a separate private\nrepository so that the leaderboard's evaluation is the single source of truth.\n\n- **Leaderboard Space**: [`HuggingAI4Engineering/CADGenBench`](https://huggingface.co/spaces/HuggingAI4Engineering/CADGenBench)\n- **Browse the tasks**: the [Tasks tab](https://huggingface.co/spaces/HuggingAI4Engineering/CADGenBench) on the Space (thumbnails, search, generation/editing filter, per-task detail)\n- **Benchmark code**: [`github.com/huggingface/cadgenbench`](https://github.com/huggingface/cadgenbench)\n- **Submissions + results**: [`HuggingAI4Engineering/cadgenbench-submissions`](https://huggingface.co/datasets/HuggingAI4Engineering/cadgenbench-submissions)\n- **Ground truth**: [`HuggingAI4Engineering/cadgenbench-data-gt`](https://huggingface.co/datasets/HuggingAI4Engineering/cadgenbench-data-gt) (private)\n\n## Dataset Summary\n\nCADGenBench contains **81 fixtures** of mechanical parts split across two\ntasks:\n\n- **Generation (49)**: reproduce the part as an accurate 3D solid from an\n  engineering drawing.\n- **Editing (32)**: apply a described change to an existing STEP solid.\n\nEach submission is one `output.step` per fixture. Outputs are scored against\nthe private ground truth by the CAD Score pipeline: a hard validity gate\nfollowed by a weighted mean of three orthogonal metrics (shape similarity,\ninterface match, topology match). See the\n[metric definitions](https://github.com/huggingface/cadgenbench/blob/main/docs/metrics.md).\n\n## Supported Tasks and Leaderboards\n\nThe benchmark is tool-agnostic: a submission is one STEP file per fixture,\nproduced by any system (one LLM, several, a script, or by hand). Submit and\nview results on the\n[leaderboard Space](https://huggingface.co/spaces/HuggingAI4Engineering/CADGenBench);\nthe full submission contract (zip layout, `meta.json`, validity gate) is in\n[`docs/benchmark/submission.md`](https://github.com/huggingface/cadgenbench/blob/main/docs/benchmark/submission.md).\n\n## Languages\n\nTask prompts and descriptions are in English.\n\n## Dataset Structure\n\n### Data Instances\n\nOne directory per fixture, named by its numeric id. There are two shapes:\n\n```\n# Generation fixture\n<id>/\n├── description.yaml      # prompt + metadata\n└── input.png            # the engineering drawing (input2.png, ... when multi-image)\n\n# Editing fixture\n<id>/\n├── description.yaml      # prompt + metadata\n├── edit_description.txt  # the requested change, as an instruction\n├── input.step           # the starting solid to edit\n├── input.mesh.npz       # trusted watertight mesh sidecar for input.step\n└── renders/             # iso / front / top / right PNGs of the starting solid\n```\n\n### Data Fields\n\n`description.yaml` carries:\n\n| Field | Type | Description |\n|---|---|---|\n| `description` | string | The task prompt. |\n| `task_type` | `\"generation\"` \\| `\"editing\"` | The task family. |\n| `input_files` | list of strings | The input files the task declares (e.g. `input.png`, `input2.png`, or `input.step`). |\n| `input_type` | `\"text+image\"` \\| `\"text+step\"` | Modality of the inputs. |\n\n### Data Splits\n\nNo train/test split — all 81 fixtures form a single evaluation set\n(49 generation, 32 editing).\n\n## Dataset Creation\n\n### Source Data\n\nThe underlying CAD geometry is sourced from [Mecado](https://www.mecado.com).\nFixtures are real mechanical parts with mating interfaces (locating jigs, bolt\npatterns, slots). For each part, an engineering drawing (generation) or a\nstart","encoding":"utf8","key":"README.md","name":"README.md","size":5057,"truncated":true}],"sourceObjectKey":"201/input.mesh.npz","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"HuggingAI4Engineering/cadgenbench-data\", filename=\"201/input.mesh.npz\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"101/input.png","totalObjects":360},"summary":"Public task inputs for CADGenBench, a benchmark of how well AI systems produce correct 3D mechanical parts as STEP solids, run by the HuggingAI4Engineering team.","title":"CADGenBench Inputs: 81 Mechanical-Part Fixtures for AI CAD Generation from Engineering Drawings and STEP Editing (Hugging Face)","updatedAt":"2026-11-22T00:44:57.803Z","visibility":"public","websiteUrl":"https://github.com/huggingface/cadgenbench"}}