SMOKE - ROS implementation

This work ports the SMOKE algorithm to a ROS1 enviroment. No changes in SMOKE architecture were made, but the weights must be loaded and then exported, in your computer.

See below the recorded SMOKE visualization on RVIz.

it is slowed down so that it is visible. But it has achieve greater perfomance (about 20fps). It was compiled with cuda 11.

first clone this repo with

git clone [email protected]:poledna/SMOKE_ros.git --recurse-submodules 

then create the virtual enviroment with:

cd SMOKE_ros/
conda env create -f smoke_pkg.yaml

activate the conda enviroment (it is importante to activate this enviroment every time before the ros source)

conda activate smoke_pkg
cd SMOKE

the build SMOKE

python setup.py build develop

with SMOKE built build the workspace

cd ../smoke_ws
catkin build 

and source the workspace

source devel/setup.bash

now you should be able to execute the smoke kernel ros code with:

cd src/smoke/src/
rosrun smoke_pkg smoke_kernel.py 

if you see

##################################################
SMOKE model loaded to GPU!

it has worked and now is waiting inputs on: /camera/image_raw. If it does not init due to file not found go to the SMOKE_ros/smoke_ws/src/smoke/src dir. (An issue Im not sure how to solve yet)

To test with a test image of kitti run:

cd src/smoke/src/
rosrun smoke_pkg send_image.py 

To view the image use:

rosrun smoke_pkg smoke_kitti_visualization.py 

and use RViz

to cite SMOKE work:

@article{liu2020SMOKE,
  title={{SMOKE}: Single-Stage Monocular 3D Object Detection via Keypoint Estimation},
  author={Zechen Liu and Zizhang Wu and Roland T\'oth},
  journal={arXiv preprint arXiv:2002.10111},
  year={2020}
}