Photorealistic training data on demand — 3.4k+ stars on GitHub
BlenderProc is a widely used open source procedural pipeline for the fast creation of photorealistic training images for deep neural networks in computer vision.
It renders RGB together with depth, segmentation, normals, and object poses, which makes it a widely used tool for reducing the reality gap in vision and robotics research — including the HOWS-CL-25 dataset built with it.
Journal of Open Source Software (JOSS), 2023
@article{denninger2023blenderproc2,
title = {BlenderProc2: A Procedural Pipeline for Photorealistic Rendering},
author = {Denninger, M. and Winkelbauer, D. and Sundermeyer, M. and Boerdijk, W. and Knauer, M. and Strobl, K. and Humt, M. and Triebel, R.},
journal = {Journal of Open Source Software (JOSS)},
year = {2023}
}
Robotics: Science and Systems (RSS) Workshops, 2020
@inproceedings{denninger2020blenderproc,
title = {BlenderProc: Reducing the Reality Gap with Photorealistic Rendering},
author = {Denninger, M. and Sundermeyer, M. and Winkelbauer, D. and Olefir, D. and Hodan, T. and Zidan Y. and Elbadrawy, M. and Knauer, M. and Katam, H. and Lodhi, A.},
booktitle = {Robotics: Science and Systems (RSS) Workshops},
year = {2020}
}