/demo-fpnet

Floorplan Recognition especially for complicated drawings

Primary LanguagePython

Deep Floor Plan Analysis for Complicated Drawings Based on Style Transfer

Seongyong Kim, Seula Park, Hyengjung Kim, Kiyun Yu
Journal of Computing in Civil Engineering, 2021.

This FPNet [Source] is able to reconstruct walls in areas with overlapping graphics or nonuniform patterns, thus allowing the room structures to be recovered even from complicated drawings.

Fig. (a) input floor plan images and our results of (b) the style-transferred plans and (c) the vectorized floor plans

Citation

@article{FPNet2020, 
    title={Deep Floor Plan Analysis for Complicated Drawings Based on StyleTransfer}, 
    author={Seongyong Kim, Seula Park, Hyeongjung Kim, Kiyun Yu}, 
    journal={Journal of Computing in Civil Engineering}, 
    year={2020}, 
    DOI={10.1061/(ASCE)CP.1943-5487.0000942}
}

Acknowledgments

Code borrows heavily from [Isola et al]. The floor plan datasets originated from EAIS-fp [Jang et al] and SNU-fp [Kim et al], and in purpose of increasing the varsarity of the datasets, we added more drawings in diverse formats.


DEMO- underworking

How to run this demo

The demo requires Python =< 3.6 (The version of TensorFlow we specify inrequirements.txt is not supported in Python 3.7+).

git clone https://github.com/streamlit/demo-face-gan.git
cd demo-fpnet
pip install -r requirements.txt
streamlit run app.py

supplementary

Fig. Results of our method for various types of complex floor plans in EAIS. (a–c) An input floor plan, our vectorized floor plan, and the corresponding 3D model. In the vectorized floor plan, the walls and openings are represented by lines, respectively.