Home

Awesome

TEGBP

Official repo for CVPR 2023 Paper "Tangentially Elongated Gaussian Belief Propagation for Event-based Incremental Optical Flow Estimation" Jun, Nagata and Yusuke Sekikawa

@inproceedings{cvpr2023tegbp,
  author    = {Jun, Nagata, and Yusuke, Sekikawa},
  title     = {Tangentially Elongated Gaussian Belief Propagation for  Event-based Incremental Optical Flow Estimation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2023},
  url={https://openreview.net/attachment?id=FppNuRVrS8&name=camera-ready_PDF}
}

normal (input) full (output)

<!-- <img src="https://github.com/DensoITLab/tegbp/blob/master/material/flo_1_00009.png" alt="normal flow" title="normal flow"> --> <!-- <img src="https://github.com/DensoITLab/tegbp/blob/master/material/flo_0_00009.png" alt="full flow" title="full flow"> -->

Indoor Flying

Setup dependency

sudo apt install libeigen3-dev
pip install Command
source /opt/intel/oneapi/setvars.sh --force

Prepare result folder

Local SSD is recomended.

ln -s /home/data2/cashe/tegbp_result/ result
sudo chmod -R 777 result
mkdir result/bricks_1slide
mkdir result/dummy
mkdir result/bricks
mkdir result/indoor_flying2

Compile

source /opt/intel/oneapi/setvars.sh
make

MakeFile options

-fexcess-precision=fast -ffast-math -mssse3 -mfpmath=sse

Run

./process #thread dataname win_size save_interval
./process =1 "indoor_flying2" -1 -1
./process -1 "bricks" 15000 // debug
 ./process -1 "indoor_flying2" -1 -1

Visualization

Run vis_result.ipynb. It'll execute the followng operations

If you can not open notebook, first convert it to the python script and then execute as a python script.

jupyter nbconvert  --to script vis_result.py  vis_result.ipynb 
python vis_result.py 

Misc

See matlab directory for the reference matlab implementation (slow).

Credit

Coder: Y.Sekikawa, J.Nagata

Reference