AI/Computer Vision for extracting dimensions and features from engineering drawings
I work in a manufacturing environment and I'm exploring whether AI/computer vision can be used to automatically interpret 2D engineering drawings.
The goal is to identify and extract:
* Components and geometric features
* Dimensions and their associated features
* Tolerances
* GD&T symbols
* Hole specifications
* Surface-finish information
* Engineering notes and annotations
Ideally, the output would be structured data that could later be used for manufacturing, inspection, costing, BOM generation, or integration with other systems.
I'm aware that OCR can extract text, but the bigger challenge seems to be understanding the **relationship between dimensions, symbols and the actual geometry/features in the drawing**.
Has anyone worked on something similar?
I'm particularly interested in:
* Vision-language models
* OCR + computer vision pipelines
* Object detection/segmentation
* Engineering drawing datasets
* CAD-aware approaches
* Open-source models or commercial APIs
What would be the most practical architecture for solving this reliably with real-world engineering drawings?