Down-sampling-based 2D floor plan segmentation

This repository contains the Python implementation for the the paper 2D Floor Plan Segmentation Based on Down-sampling. This algorithm performs segmentation on a 2D floor plan by using a down-sampling technique. The overall pipeline is shown below. For more details please refer to the paper.

Detailed comparison with the state-of-the-art methods

Click on any of the following cases to view the comparison results with the three state-of-the-art methods introduced in Bormann et al. (2016). From left to right the methods are: (1) ground truth labels, (2) morphologic segmentation, (3) distance-transform-based segmentation, (4) Voronoi-graph-based segmentation, and (5) our down-sampling-based segmentation.

With furniture

No furniture

Usage

For quick testing of the package we recommend using Docker. We have tested the code on Ubuntu 20.04. First install the latest version of docker. Then, follow the following instructions:

Docker container usage

# Build the docker image
docker build --build-arg USERID=$(id -u) --build-arg GROUPID=$(id -g) -t py_floor_plan_segmenter .

# Running the container in interactive mode
docker run -u cloud -v .:/app -it py_floor_plan_segmenter bash

# Example command to perform segmentation on the rank.png image in the `/app/sandbox/maps/sample_1` folder.
python -m py_floor_plan_segmenter -i /app/sandbox/maps/sample_1 -p /app/sandbox/out
# Example command with debug flag to generate debugging images
python -m py_floor_plan_segmenter -i /app/sandbox/maps/sample_1 -p /app/sandbox/out --debug
# Example command with debug and animate flags to generate debugging images and animation
python -m py_floor_plan_segmenter -i /app/sandbox/maps/sample_1 -p /app/sandbox/out --debug --animate

Advanced usage: Running the API server

To run the API server:

# Clone the repository
git clone https://gitlab.tmecosys.net/mohammadreza.sharif/py_room_segmenter
cd py_room_segmenter

docker-compose build --build-arg USERID=$(id -u) --build-arg GROUPID=$(id -g)
# To start the server
docker-compose up -d
# To stop the server
docker-compose stop

Then, open your browser at the address for documentation: http://0.0.0.0:8008/docs

Running the benchmarks and tests

In docker interactive mode:

cd /app

# Run benchmark furnished maps
./run_benchmark.sh sandbox/maps/benchmark/furnished

# Run benchmark maps with no furniture
./run_benchmark.sh sandbox/maps/benchmark/no_furniture

# Run Neato dataset present in the paper
./run_benchmark.sh sandbox/maps/neato/paper

# Run Neato extra dataset
./run_benchmark.sh sandbox/maps/neato/extra

Evaluation of benchmarks

# Evaluate the furnished maps benchmark
./run_evaluation.sh sandbox/out/furnished sandbox/maps/benchmark/groundtruth

# Evaluate the no-furniture maps benchmark
./run_evaluation.sh sandbox/out/no_furniture sandbox/maps/benchmark/groundtruth

Citation

Please use the following BibTeX entry for citing this work:

@ARTICLE{Sharif2023-bt,
  title         = "{2D} Floor Plan Segmentation Based on Down-sampling",
  author        = "Sharif, Mohammadreza and Mohan, Kiran and Suvarna, Sarath",
  month         =  mar,
  year          =  2023,
  archivePrefix = "arXiv",
  primaryClass  = "cs.CV",
  eprint        = "2303.13798"
}