Rectified Mask Extractor of Plan2Scene

We use the following 3rd party projects which are imported as git submodules. Please agree to their terms of use as well if you use this code.

1. https://github.com/CSAILVision/semantic-segmentation-pytorch
2. https://github.com/svip-lab/PlanarReconstruction

How to use?

1. Download the pre-trained checkpoints for 'ade20k-hrnetv2-c1' available here. There are two checkpoint files 'encoder_epoch_30.pth' and 'decoder_epoch_30.pth'.

2. Download the pre-trained model for PlanarReconstruction provided by the authors.

3. Install pre-requisites of PlanarReconstruction and semantic-segmentation-pytorch projects to a common python environement. You can use the provided environment.yml file to create a conda environment that has dependencies of both projects.

   conda env create -f environment.yml
   conda activate plane

4. Rename ./conf/data-paths-example.json to ./conf/data-paths.json. Update checkpoint path fields of it.

5. Extract rectified surface masks as follows.

   export PYTHONPATH=./code/src
   python ./code/scripts/extract_surface_masks.py ./data/output/floor /path/to/rent3d/dataset/images/ floor
   python ./code/scripts/extract_surface_masks.py ./data/output/wall /path/to/rent3d/dataset/images/ wall
   python ./code/scripts/extract_surface_masks.py ./data/output/ceiling /path/to/rent3d/dataset/images/ ceiling

   The rectified surface masks are generated at ./data/output/[SURFACE]/rectified_surface_masks directory. You can open ./data/output/[SURFACE]/index.html to preview the extracted surface masks.cdcdcdcd