data_preprocess/README.md

Data preprocessing

Structured3D

Simply download preprocessed data by RoomFormer at here. For more details, please refer to RoomFormer's instructions.

To render binary floorplan images from GT annotations (as used in our paper), run bash data_preprocess/tools/run_s3d.sh.

CubiCasa5K

Step 1: Download and extract CubiCasa5K dataset.

Step 2: Run bash data_preprocess/cubicasa5k/run.sh.

Raster2Graph

The instruction mainly follows Raster2Graph's instruction.

Step 1: Due to dataset proprietary restrictions, please apply for access to LIFULL HOME'S Data here.

Step 2: After obtaining access, download only the "photo-rent-madori-full-00" folder, which contains approximately 300,000 images.

Step 3: Apply for access to the annotation here.

The package has 3 folders:

• annot_npy, annot_json: the annotations saved in npy and json, respectively.
• original_vector_boundary: boundary boxes of "LIFULL HOME'S Data" which is used to create centered 512x512 images.

These folders should be saved in the same directory as photo-rent-madori-full-00. For example: data/R2G_hr_dataset/.

Step 4: Run bash data_preprocess/tools/run_r2g.sh.

WAFFLE

It is noted that since WAFFLE only provides segmentation masks for a subset of 100 examples, so we only process this subset for the evaluation, not for training.

Step 1: Download data at here

Step 2: Run bash data_preprocess/tools/run_waffle.sh.

Data visualization

Please refer to this script tools/plot_data.sh.