This library includes code to generate the synthetic training set, OWID, which is for VoxDet training.

You can easily customize your own instance dataset using this toolkit for higher-quality 3D CAD models!

The code consists of two parts Render_OWID is a set of tool scripts that use the functions in BlenderProc to load 3D models and render images. BlenderProc is a modified version of the public BlenderProc2 library to better fit our needs

Please follow the steps to render the dataset.

For reference, this code is built upon the tutorial here, you can also have a look for deeper understanding.

• ShapeNet, we used the ShapeNetCore_v1 dataset.
• ABO, we used the above-3dmodels.tar
• cc_texture, please follow the description in the tutorial

cd BlenderProc
pip install -e .

cd Render_OWID
bash render_p1_0.sh # please modify the directions according to your ABO and ShapeNet path

cd train_set
python3 format_p1.py # please modify the directions accordingly

This formats p1 data into a structured folder and also generates the mapping dictionary for p2 rendering.

cd ..
bash render_p2_0.sh # you can open 4 tmux windows and run the render_p2_0,1,2,3.sh in parallel, please modify the directions accordingly

cd train_set
python3 format_p2.py # please modify the directions accordingly
python3 split_p2.py # please modify the directions accordingly, you can also set a custom train/val ratio

Note: In Render_OWID/test_set, we also provide the script to render a 360 video of the test instances, it follows a similar procedure for the training set.

The authors sincerely thank the developers of Blenderproc2, we build our toolkit upon their code library and tutorial. The authors would also like to thank the authors of ShapeNet and ABO for providing high-quality 3D models.

If our toolkit helps your research, please cite us as:

@INPROCEEDINGS{Li2023Vox,       
	author={Li, Bowen and Wang, Jiashun and Hu, Yaoyu and Wang, Chen and Scherer, Sebastian},   
	booktitle={Proceedings of the Advances in Neural Information Processing Systems (NeurIPS)}, 
	title={{VoxDet: Voxel Learning for Novel Instance Detection}},
	year={2023},
	volume={},
	number={}
}