cad-environments

# CAD Environments CAD Environments is a multimodal dataset of complete, human-performed workflows in desktop CAD software. The current release contains **51 task workflows** totaling **99.03 hours**, covering eight software groups across mechanical design, architecture, MEP, structural design, and general 3D modeling. Each workflow preserves the full task context—not just the final model—including the problem statement, reference and input files, a gold output, evaluation rubrics, a human walkthrough screen recording, synchronized interaction events, frame timing, and narration. ## Dataset summary | Metric | Value | |---|---:| | Tasks / workflows | 51 | | Human walkthrough recording time | 99.03 hours | | Software groups | 8 | ## Software breakdown | Software | Tasks / workflows | Hours | |---|---:|---:| | AutoCAD | 4 | 25.70 | | SketchUp | 7 | 25.15 | | SOLIDWORKS | 19 | 16.64 | | Revit Architecture | 6 | 11.70 | | Revit MEP | 1 | 7.19 | | Siemens NX | 6 | 6.25 | | Revit Structure | 3 | 4.26 | | STAAD.Pro | 5 | 2.14 | | **Total** | **51** | **99.03** | ## What each workflow contains - **Problem statement:** `task_desc.json` provides the structured task, application, domain, description, deliverables, and any specified dimensions. - **Human-readable task overview:** `task_overview.pdf` packages the task for convenient review. - **Reference and source material:** `input_files/` contains the files and assets provided to the operator. - **Gold output:** `output_files/` contains the completed CAD deliverables produced during the walkthrough. - **Evaluation criteria:** `rubrics.json` describes the requirements used to judge task completion. - **Human walkthrough:** `clip.mp4` records the complete on-screen execution of the task. - **Interaction trace:** `events.json` contains synchronized mouse and keyboard events; `frame_events.json` provides frame-level timing and event alignment. - **Narration:** `narration.json` provides timestamped descriptions of the workflow. - **Recording metadata:** `metadata.json` includes duration, frame rate, frame count, event count, platform, and category. - **Derived computer-use representation:** `scale-cua-format/` contains trajectory annotations and step images for frame/action-based use. ## Directory structure `data/workflows.parquet` is a 51-row workflow index used by the Hugging Face Dataset Viewer. The raw contents of each workflow follow this structure: ```text <software>/<workflow-id>/ ├── task_desc.json ├── task_overview.pdf ├── rubrics.json ├── metadata.json ├── clip.mp4 ├── events.json ├── frame_events.json ├── narration.json ├── input_files/ ├── output_files/ └── scale-cua-format/ ├── annotations/trajectory.json └── data/images/ ``` The canonical software prefixes are `autocad/`, `nx-cad/`, `revit-architecture/`, `revit-mep/`, `revit-structure/`, `sketchup/`, `solidworks/`, and `staad-pro/`.