3DB: A framework for analyzing computer vision models with simulated data
Installation
Follow instructions on: https://github.com/3db/installers
Complete demo
For detailed, step by step demonstration of the usage of the framework, please visit: https://github.com/3db/demo
Documentation
You can find in-depth documentation for the package here, including a quickstart guide, an explanation on the internal layout of the package, as well as guides for extending 3DB to new modalities, tasks, and output formats.
Primer data
We offer data for users to quickly get started 3DB and reproduce the results of the paper mentioned below. It is available on dropbox at this link: https://www.dropbox.com/s/2gdprhp8jvku4zf/threedb_starting_kit.tar.gz?dl=0. (One can use wget
to download it).
It contains:
- 3D models
- Environments: our studio and HDRI backgrounds
- Replacement textures to use with
threedb.controls.blender.material
- Licensing/credits for the data
Citation
If you find 3DB helpful, please cite it as:
@inproceedings{leclerc2021three,
title={3DB: A Framework for Debugging Computer Vision Models},
author={Guillaume Leclerc, Hadi Salman, Andrew Ilyas, Sai Vemprala, Logan Engstrom, Vibhav Vineet, Kai Xiao, Pengchuan Zhang, Shibani Santurkar, Greg Yang, Ashish Kapoor, Aleksander Madry},
year={2021},
booktitle={Arxiv preprint arXiv:2106.03805}
}