Turning Cabinet Drawings into 3D Models with AI

Source: Dev.to

Introduction

Construction and cabinet manufacturing still rely heavily on PDF drawings created by designers. The process is slow and repetitive. This project asks a simple question: Can AI convert cabinet drawings directly into usable 3D data?

We built a system that reads cabinet drawings from PDFs, extracts structured geometry, and generates DWG and 3D models.

Why Traditional Automation Falls Short

Cabinet drawings contain valuable information—layout structure, cabinet boundaries, measurements, labels, door and drawer positions—but most of it exists only in visual form. Traditional automation tools expect structured CAD data, not messy PDFs, making it difficult for machines to interpret the drawings.

Pipeline Overview

The workflow combines computer vision, detection models, OCR, and language models:

1. PDF → Images – Convert each page of the PDF into raster images.
2. Object Detection – Use a YOLO‑based model to detect cabinets and components.
3. Text Extraction – Apply OCR to capture measurement text.
4. LLM Interpretation – Convert ambiguous measurements into structured data.
5. Geometry Generation – Build parametric cabinet objects.
6. Output Production – Export DWG files, 3D assemblies, and layout visualizations.

Each step solves a specific problem in the conversion pipeline.

Object Detection with YOLO

We trained a YOLO‑based detector to identify the following components inside drawings:

• Base cabinets
• Wall cabinets
• Tall cabinets
• Appliances
• Structural boundaries

Why YOLO?
YOLO provides fast detection with high spatial accuracy, which is critical for architectural drawings where precise bounding boxes are required.

After detection, the system extracts bounding boxes and spatial relationships, forming the foundation for geometry reconstruction.

Measurement Extraction

Cabinet drawings include measurements such as width, height, depth, and spacing. OCR pipelines pull the raw text, which often appears in inconsistent formats, e.g.:

W 36"
H 34 1/2"
D 24"

LLM Interpretation

The raw OCR output is passed to a large language model (LLM) that normalizes the data into a structured format.

Example conversion

Raw text

36 W x 34.5 H x 24 D

Structured JSON

{
  "width": 36,
  "height": 34.5,
  "depth": 24
}

The LLM also resolves:

• Inconsistent labels
• Missing context
• Varied measurement formatting

This step turns visual annotations into reliable numerical data.

Geometry Generation

With cabinet detections, dimensions, and layout relationships, we generate parametric objects:

{
  "type": "Base",
  "width": 36,
  "height": 34.5,
  "depth": 24,
  "position": {"x": 0, "y": 0}
}

From this structure we can produce:

• 3D models
• AutoCAD DWG files
• Manufacturing layouts

Designers can open the results directly in CAD software, eliminating hours of manual drafting.

Real‑World Challenges

• Variability – No two cabinet drawings are identical; annotation styles, measurement formats, and symbols differ widely.
• Scale Conversion – Architectural drawings use scaled representations; we must translate pixel distances into real‑world dimensions.
• Context Awareness – Cabinets interact with walls, appliances, and adjacent units, requiring layout context beyond isolated object detection.

Even small errors can break cabinet assembly, so robustness is essential.

Benefits and Future Directions

Automating cabinet interpretation unlocks:

• Faster cabinet design workflows
• Automated CAD generation
• Reduced manual drafting effort
• Accelerated manufacturing preparation

Future work aims to process entire architectural plans, not just cabinets. Many industries still rely on human‑focused documents; combining computer vision, detection models, and language models can convert visual design documents into structured data pipelines. Cabinet drawings are just one example of this broader opportunity.