mirror of
https://github.com/karma-riuk/hdr_esim.git
synced 2024-11-23 14:47:50 +01:00
Super HDR reconstruction with event cameras. Extending ESIM generate more datasets to train NeRFs with
event_camera_simulator | ||
dependencies.yaml | ||
README.md |
ESIM: an Open Event Camera Simulator
This is the code for the 2018 CoRL paper ESIM: an Open Event Camera Simulator by Henri Rebecq, Daniel Gehrig and Davide Scaramuzza:
@Article{Rebecq18corl,
author = {Henri Rebecq and Daniel Gehrig and Davide Scaramuzza},
title = {{ESIM}: an Open Event Camera Simulator},
journal = {Conf. on Robotics Learning (CoRL)},
year = 2018,
month = oct
}
You can find a pdf of the paper here. If you use any of this code, please cite this publication.
Features
- Accurate event simulation, guaranteed by the tight integration between the rendering engine and the event simulator
- IMU simulation
- Support for multi-camera systems
- Ground truth camera poses, IMU biases, angular/linear velocities, depth maps, and optic flow maps
- Support for camera distortion
- Different C+/C- contrast thresholds
- Basic noise simulation for event cameras (based on additive Gaussian noise on the contrast threshold)
- Motion blur simulation
- Publish to ROS and/or save data to rosbag
Install
Installation instructions can be found in our wiki.
Run
Specific instructions to run the simulator depending on the chosen rendering engine can be found in our wiki.
Acknowledgements
We thank Raffael Theiler and Dario Brescianini for their contributions to ESIM. This research was supported by by Swiss National Center of Competence Research Robotics (NCCR), Qualcomm (through the Qualcomm Innovation Fellowship Award 2018), the SNSF-ERC Starting Grant and DARPA FLA.