CADExpert

# CADExpert Dataset ## Overview CADExpert is a large-scale benchmark dataset designed for **executable and editable CAD code generation from engineering drawings**. It addresses the lack of high-quality datasets that align with real-world industrial design workflows. The dataset contains high-fidelity samples, each consisting of orthographic projection drawings with precise dimension annotations, expert reasoning processes, executable CAD programs, and corresponding 3D models. CADExpert aims to support research on multimodal reasoning, programmatic CAD modeling, and vision-language based engineering design. ## Core Features 1. **Engineering-Drawing Inputs**: Each sample uses orthographic projection drawings with precise dimension annotations, closely reflecting real engineering design workflows. 2. **Executable CAD Programs**: The dataset provides fully executable CadQuery (Python-based) scripts that generate parametric CAD models. 3. **Expert Reasoning Processes**: Each sample includes step-by-step reasoning (Chain-of-Thought) describing how the CAD model is constructed from the drawings. 4. **Industrial-Level Modeling Operations**: The dataset includes complex CAD operations such as chamfering, drilling, boolean difference, and polyline trajectories, enabling realistic industrial modeling scenarios. ## Sample Structure Each sample contains the following components: | Field | Type | Description | |------|------|-------------| | `orthographic_views` | Image | Orthographic projection drawings with dimension annotations | | `reasoning_cot` | String | Expert reasoning process for CAD code generation | | `cadquery_code` | String | Executable CadQuery script for parametric modeling | | `id` | String | Unique identifier | ## Use Cases - Training multimodal models for **engineering drawing to CAD code generation** - Benchmarking CAD generation models using **geometry-based metrics** - Research on **vision-language reasoning for parametric modeling** - Development of **AI-assisted CAD design systems**