BoopNet

A U-Net-inspired convolutional encoder-decoder to extract floor plans from commercial real estate brochures in the form of PDF documents.

Version

Both the package and the model are at v0.1.

Requirements

• imagemagick
• pip

Usage

from boopnet import extract_from_pdf

extract_from_pdf("/home/boop/stuff/brochure.pdf")

This will yield one PNG file for each page where a floor plan ( or more ) are found. Floor plan images are tightly cropped.

You may pass destination="/home/boop/floor_plans" to this function should you wish for the PNGs to be exported into a different directory.

from boopnet import extract_all_from_directory

extract_all_from_directory("/home/stuff/pdfs")

This will work through all PDFs in the specified directory. Again, you may specify a destination directory if you want.

Installation

ImageMagick

sudo apt-get install imagemagick

or

brew install imagemagick

Python dependencies

pip install -r requirements.txt

To Do

• For pages containing multiple, distinct floor plans, export these as individual images
• Allow user to specify resolution, DPI, and quality of image to pull from the PDFs
• Train the model on more and better data to improve result quality
• Create a setup.py file

Authors

Brittney Johnson and Quentin Caudron