Indoor Layout Estimation from a Single Image

TODO

• Upload trained model.
• Upload test results.

Original readme

Original paper

Update (May 7th, 2019)

• A trained model and LSUN dataset can be downloaded from my google drive page (access permission required).

Current best training practice

• Command

  python main.py --phase train --name train --l2_factor 5  --edge_factor 50

Usage

• Install libraries according to requirements.

• Download dataset from http://lsun.cs.princeton.edu/2015.html#layout and put them in following folders.

• Datasets preparation

  • 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
  • Run the following to prepare train/evaluation datasets.

    python re_label.py

• Training

  • The trained model will be saved to folder ./exp/checkpoints/
  • You can modify config.yml to play with hyperparameters for training.

  python main.py --phase train --name train
  
  usage: main.py [-h] [--epoch EPOCH] [--batch_size BATCH_SIZE] [--lr LR]
                 [--worker WORKER] [--name NAME] [--folder FOLDER]
                 [--dataset {lsunroom,hedau,sunrgbd}]
                 [--phase {train,eval,eval_search}] [--image_size IMAGE_SIZE]
                 [--use_edge] [--use_corner] [--datafold DATAFOLD]
                 [--tri_visual] [--arch ARCH] [--optim OPTIM] [--disjoint_class]
                 [--pretrain_path PRETRAIN_PATH] [--l1_factor L1_FACTOR]
                 [--l2_factor L2_FACTOR] [--edge_factor EDGE_FACTOR]
                 [--focal_gamma FOCAL_GAMMA]
  
  Indoor room corner detection
  
  optional arguments:
    -h, --help            show this help message and exit
    --epoch EPOCH
    --batch_size BATCH_SIZE
    --lr LR
    --worker WORKER
    --name NAME           experiment name
    --folder FOLDER       where is the dataset
    --dataset {lsunroom,hedau,sunrgbd}
    --phase {train,eval,eval_search}
    --image_size IMAGE_SIZE
    --use_edge
    --use_corner
    --datafold DATAFOLD
    --tri_visual
    --arch ARCH
    --optim OPTIM
    --disjoint_class
    --pretrain_path PRETRAIN_PATH
    --l1_factor L1_FACTOR
    --l2_factor L2_FACTOR
    --edge_factor EDGE_FACTOR
    --focal_gamma FOCAL_GAMMA

• Prediction

  • Specify the weight path of a trained model.
  • The weight path should be a file named as net-xx.pt
  • --input_path/--output_path point to the folders of input/output images.
  • A notebook to visualize the prediction result

  python demo.py --weight [weight_path] --input_path [input_image_dir] --output_path [output_image_dir]
  
  Usage: demo.py [OPTIONS]
  
  Options:
    --input_path PATH
    --output_path PATH
    --weight PATH
    --input_size <INTEGER INTEGER>...
    --help                          Show this message and exit.
  

MIT License