---
title: "From 2D Drawings to 3D Models: Designing an AI Reconstruction Pipeline on MongoDB"
description: "BIRD is a research framework that converts five 2D perspective drawings into a proportionate, dimensioned 3D building envelope. It uses a dual-branch CNN to separate architectural line segments from noise, a text recognition model to scale them to real-world inches, and a wrapping algorithm to fold the elevation views around the floor plan. The framework achieved a median accuracy of roughly 93% across complete building plan sets from small residences to a multi-story apartment complex. The proposed deployment on MongoDB Atlas stores each project as a single JSON document containing all drawings’ segments, corrections, and final geometry, with S3 for image storage and Atlas Stream Processing for event-driven workflow coordination. MongoDB Search and Vector Search with Voyage AI embeddings enable full-text and semantic search over past projects, allowing drafters to start from corrected models of similar builds."
---

# From 2D Drawings to 3D Models: Designing an AI Reconstruction Pipeline on MongoDB

[MongoDB](https://yomu.fyi/company/mongodb) · Sep 22, 2026

**Type:** Explainer

## Summary

BIRD is a research framework that converts five 2D perspective drawings into a proportionate, dimensioned 3D building envelope. It uses a dual-branch CNN to separate architectural line segments from noise, a text recognition model to scale them to real-world inches, and a wrapping algorithm to fold the elevation views around the floor plan. The framework achieved a median accuracy of roughly 93% across complete building plan sets from small residences to a multi-story apartment complex. The proposed deployment on MongoDB Atlas stores each project as a single JSON document containing all drawings’ segments, corrections, and final geometry, with S3 for image storage and Atlas Stream Processing for event-driven workflow coordination. MongoDB Search and Vector Search with Voyage AI embeddings enable full-text and semantic search over past projects, allowing drafters to start from corrected models of similar builds.

## Context

Small and mid-size building manufacturers and renovation firms often need 3D models from 2D drawings for quoting, visualization, or planning, but currently rebuild geometry manually in CAD tools, leading to rework and delays. BIRD automates this conversion.

## Approach / What changed

BIRD uses a dual-branch CNN to extract clean line segments from noisy drawings, a text recognition model to scale segments to real-world dimensions, and a wrapping algorithm to fold elevation views around the floor plan into a 3D envelope. The proposed deployment on MongoDB Atlas uses S3 for image storage, Atlas Stream Processing for event-driven pipeline coordination, and MongoDB Search and Vector Search with Voyage AI embeddings for project search.

## Takeaways

- BIRD achieves a median accuracy of roughly 93% in visually aligning with expected final visualizations across complete building plan sets.
- The dual-branch CNN subtracts a noise heatmap from a line segment heatmap to produce clean architectural segments.
- MongoDB’s document model allows storing all five drawings’ segments, corrections, and final geometry in a single project document, avoiding join-heavy queries.

## FAQ

### How does BIRD handle noise in building plans?

BIRD uses a dual-branch convolutional neural network: one branch produces a heatmap of genuine architectural line segments, the other produces a heatmap of noise. Subtracting the noise map from the line map leaves clean segments.

### What accuracy does BIRD achieve?

Across complete building plan sets from small residences to a multi-story apartment complex, BIRD achieved a median accuracy of roughly 93% in terms of visually aligning with the expected final visualization.

### How does the MongoDB architecture enable search across past projects?

MongoDB Search provides full-text search over project metadata and notes, while MongoDB Vector Search with Voyage AI embeddings enables semantic search over drawing images and review notes, allowing drafters to find geometrically similar past projects.

**Tags:** [AI](https://yomu.fyi/topic/ai), [Machine Learning](https://yomu.fyi/topic/machine-learning), [MongoDB Atlas](https://yomu.fyi/topic/mongodb-atlas)

- Source: [MongoDB](https://www.mongodb.com/company/blog/innovation/from-2d-drawings-3d-models-designing-ai-reconstruction-pipeline)
- Source URL: https://www.mongodb.com/company/blog/innovation/from-2d-drawings-3d-models-designing-ai-reconstruction-pipeline
- Ingested by Yomu: 2026-09-23T15:00:14.794Z

[Read original post](https://www.mongodb.com/company/blog/innovation/from-2d-drawings-3d-models-designing-ai-reconstruction-pipeline)