Indoor Layout Estimation from a Single Image

Citation

@inproceedings{lin2018layoutestimation,
    Author = {Hung Jin Lin and Sheng-Wei Huang and Shang-Hong Lai and Chen-Kuo Chiang},
    Title = {Indoor Scene Layout Estimation from a Single Image},
    Booktitle = {2018 24th International Conference on Pattern Recognition (ICPR)},
    Year = {2018}
}

The code is under evaluation and update TBD. Deprecated information below.

Prerequisite

• Python 3.6+
• OneGAN >= 0.3.0
• scikit-image and click, tqdm

Usage

• Dataset

  • Put LSUN Room Layout Dataset in folder ../data/lsun_room relative to this project.

    • images/: RGB color image *.jpg of indoor room scene
    • layout_seg/: layout ground truth *.mat of indoor room scene
    • layout_seg_images/: generated layout ground truth *.png of indoor room scene

  • Put SUN RGB-D Dataset in folder ../data/sun_rgbd relative to this project.

    • images/: RGB color image *.jpg in train and test respectly.
    • labels/: layout ground truth *.png in train and test respectly.

• Toolkit

  • Put LSUN Room Layout Dataset toolkit in folder ../lsun_toolkit
    • Integrated scripts (TBD)

• Training

  python main.py
  
  Usage: main.py [OPTIONS]
  
  Options:
    --name TEXT
    --dataset [lsun_room | others]
    --dataset_root TEXT
    --log_dir TEXT
    --image_size <INTEGER INTEGER>
    --epochs INTEGER
    --batch_size INTEGER
    --workers INTEGER
    --l1_weight FLOAT
    --resume PATH

• Demo

  python demo.py
  
  Usage: demo.py [OPTIONS]
  
  Options:
    --device INTEGER
    --video TEXT
    --weight TEXT
    --input_size <INTEGER INTEGER>.
  

• Evaluate with offical Matlab toolkit

  matlab -nojvm -nodisplay -nosplash -r "demo('$EXPERIMENT_OUTPUT_FODLER'); exit;"

Tools

• Re-label

  • Output layout image (range from 1-5)

  python script/re_label.py