B-Pose: Bayesian Deep Network for Camera 6-DoF Pose Estimation from RGB Images

Here, we introduce, B-Pose, a Bayesian Convolutional deep network capable of not only automatically estimating the camera's pose parameters from a single RGB image but also providing a measure of uncertainty for the estimated parameters. BPose is published in IEEE Robotics and Automation Letters

P.S. This code is built on top of the DSAC* repository.

Introduction

• BPose takes an RGB image (and optional 3D point cloud) and returns pose parameters of the camera with associated uncertainty.
• BPose adopts the DSAC* method as the core end-to-end module and uses Bayesian Neural Nets in the last layers to make a Bayesian prediction.
• The Samling is performed on the weights of the last layers and several scene coordinate estimations have been obtained.
• The pose estimation and refinement have been performed for all scene coordinates.
• The final pose is the mean of all estimated poses and the uncertainty is the associated variance of estimations.

Installation

Since this code is built on DSAC* code, Bpose requires the same python packages as DSAC* required. Additionally, we use Pyro for implementing our Bayesian network.

pip install pyro-ppl

Citations

If you found this page helpful, please cite the following survey papers:

@article{rekavandi2023b,
  title={B-Pose: Bayesian Deep Network for Camera 6-DoF Pose Estimation from RGB Images},
  author={Rekavandi, Aref Miri and Boussaid, Farid and Seghouane, Abd-Krim and Bennamoun, Mohammed},
  journal={IEEE Robotics and Automation Letters},
  year={2023},
  publisher={IEEE}
}