Run bash prepare_data.sh to download all the datasets used in this project
and generate the multi-view renders for shape representation.
-
This could take a very long time as there are 6 datasets to download and a large amount of images to generate.
-
To accelerate the data preparation, you can run the bash file for one dataset at a time.
To generate point clouds for Pascal3D and ObjectNet3D.
First make sure that you have generated the correct .obj files by running
python off2obj.py
Then you need to get the dependencies for the virtual_canner:
apt-get install -y --no-install-recommends libboost-all-dev libcgal-dev libeigen3-dev
and build it:
cd virtual_scanner
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make
You should find the executable_virtual_scanner in ./virtual_scanner/build/virtualscanner
- To generate point cloud for Pascal3D, run:
python point_cloud.py --dataset_dir Pascal3D --dataset_format Pascal3D --input CAD --virtualscanner #Path_of_the_executable_virtual_scanner
- To generate point cloud for ObjectNet3D, run:
python point_cloud.py --dataset_dir ObjectNet3D --dataset_format Pascal3D --input CAD/obj --virtualscanner #Path_of_the_executable_virtual_scanner