MLSTRUCT-FP_benchmarks

Benchmarks of MLSTRUCT-FP dataset.

Description

This repo contains the segmentation and vectorization models for processing our MLSTRUCT-FP dataset. See the following jupyter notebook files for more information and a quick start (in order):

• create_data: Creates a dataset, assembles crops, and export data session
• fp_unet: Creates U-Net model for wall segmentation
• vectorization: Vectorizes a model using Egiazarian et al. method

The weights for the best model (no_rot_256_50) can be downloaded at this link. For the vectorization model, follow the following links to download the weights for model_curves and model_lines; check the vectorization original repo for more details.

First steps

To use this code, you need a Python 3.8 installation. Then, run the following setup (assuming conda manager):

# Clone this repo
git clone https://github.com/MLSTRUCT/MLSTRUCT-FP_benchmarks.git
cd MLSTRUCT-FP_benchmarks

# Create conda environment & install deps.
conda create -n mlstructfp python=3.8
conda activate mlstructfp
pip install -e .
conda install jupyter

# Run the notebook
jupyter notebook

# Optional: Install CUDA toolkit if Tensorflow cannot detect GPU
conda install cudatoolkit=10.1 cudnn=7.6 -c conda-forge

CUDA 10.1 + cuDNN 7.6.5 is required to run the trained models.

Citing

@article{Pizarro2023,
  title = {Large-scale multi-unit floor plan dataset for architectural plan analysis and
           recognition},
  journal = {Automation in Construction},
  volume = {156},
  pages = {105132},
  year = {2023},
  issn = {0926-5805},
  doi = {https://doi.org/10.1016/j.autcon.2023.105132},
  url = {https://www.sciencedirect.com/science/article/pii/S0926580523003928},
  author = {Pablo N. Pizarro and Nancy Hitschfeld and Ivan Sipiran}
}

Author

Pablo Pizarro R. | 2023 - 2025