We have used SUNCG for the data generation, be aware that SUNCG is no longer publicly available. So you can only progress here, if you already have a downloaded version of SUNCG. Be aware that, we are not at liberty to release the assets ourselves.
Assuming you have this, we use the SUNCG Toolbox: https://github.com/tinytangent/SUNCGtoolbox
So, please follow their instructions on how to download and build it.
Afterwards, you can convert some house.json files, which you can find in the house folder. Where each hash id in this folder is a certain house.
cd house/10704e82d0ef2bf37a658af4fb81c06c
{INSTALL_FOLDER}/SUNCGtoolbox/gaps/bin/x86_64/scn2scn house.json house.obj
{INSTALL_FOLDER}/SUNCGtoolbox/gaps/bin/x86_64/scn2cam house.json camerapositions
You will need the house.obj file and the camera positions for the SDFGen and also for the generation of the color & normal images.
This process has to be done for all house.json
files, which you want to use for the training or testing.
We offer here a script to do this for you automatically, see here
Please change these two paths:
suncg_toolbox_bin_path = "/home/max/workspace/SUNCGtoolbox/gaps/bin/x86_64"
suncg_house_folder = "/home/max/Downloads/version_1.1.0/house"
The house.obj and camera position files are then stored in a data folder, which is automatically generated. With the following structure:
data
feeb4209a2f65807d3a2288b85f44a0e <-- this is the house id
obj
house.obj <-- the generated house.obj file
cam
camerapositions <-- the generated camerapositions file