Extending LibRS's GL support to RS ROS2 #2956
Conversation
Arun-Prasad-V
force-pushed
the
gl_support
branch
from
04d3df4
to
920e708
Compare
realsense2_camera/CMakeLists.txt
Outdated
| @@ -117,6 +117,7 @@ find_package(tf2 REQUIRED) | |||
| find_package(diagnostic_updater REQUIRED) | |||
|
|
|||
| find_package(realsense2 2.54.1) | |||
| find_package(realsense2-gl 2.54.1) | |||
There was a problem hiding this comment.
This means we will need to publish realsense2-gl as well right?
We need to think about if we need to make this optional somehow
There was a problem hiding this comment.
Have added a cmake variable BUILD_ACCELERATE_WITH_GPU.
There was a problem hiding this comment.
As discussed, can you check how we do it in the viewer?
Do we depend on realsense-gl and can disable/enable on runtime?
There was a problem hiding this comment.
As discussed, can you check how we do it in the viewer? Do we depend on realsense-gl and can disable/enable on runtime?
Yes, I think we should release/publish realsense2-gl as well.
The viewer is depending on realsense2-gl, and you can choose this from the settings of the viewer, whether to run or not the GLSL.
@Arun-Prasad-V, did you test if user changes the parameter from 0 to 1 during runtime, while the build was without OpenGL ?
There was a problem hiding this comment.
We need to follow this guidelines same as for the SDK.
- Users w/o GPU are not affected (meaning users w/o GPU can install and run the ROS node)
- Users with GPU can decide if to leverage the GPU or not (launch file is sufficient)
- We need to activate the internal formal conversion using GPU as well
There was a problem hiding this comment.
@Arun-Prasad-V, did you test if user changes the parameter from 0 to 1 during runtime, while the build was without OpenGL ?
Right now, dynamic change is not supported. User must set this param during launch.
And, if the build was without OpenGL, then this param won't be exposed to user.
realsense2_camera/CMakeLists.txt
Outdated
| @@ -215,6 +218,7 @@ set(dependencies | |||
| nav_msgs | |||
| tf2 | |||
| realsense2 | |||
| realsense2-gl | |||
There was a problem hiding this comment.
Same, OK for POC
For final solution we will need to rethink about it
There was a problem hiding this comment.
@Nir-Az I don't think there is another way for doing that.. except if you want to create new branch for ROS, let's say, ros2-development-gl :)
There was a problem hiding this comment.
OK, So we will have to make sure this compile on platforms w/o a GPU, and that ROS build farm accept it
realsense2_camera/src/parameters.cpp
Outdated
| @@ -82,6 +82,10 @@ void BaseRealSenseNode::getParameters() | |||
| _base_frame_id = _parameters->setParam<std::string>(param_name, DEFAULT_BASE_FRAME_ID); | |||
| _base_frame_id = (static_cast<std::ostringstream&&>(std::ostringstream() << _camera_name << "_" << _base_frame_id)).str(); | |||
| _parameters_names.push_back(param_name); | |||
|
|
|||
| param_name = std::string("data_processing_mode"); | |||
There was a problem hiding this comment.
maybe accelerate_with_gpu , on/off``?
Or as is with values CPU/GPU ?
There was a problem hiding this comment.
Changed the param name to accelerate_with_gpu and declared this param as int not bool. Because, if in case in future, we need to add new values (like for CUDA, etc.,), it can be added easily to the int param.
@Nir-Az, your thoughts?
There was a problem hiding this comment.
Not sure CUDA is replacing GPU, let's make sure how it works in LibRS
Arun-Prasad-V
force-pushed
the
gl_support
branch
from
920e708
to
da05add
Compare
| @@ -22,6 +22,10 @@ using namespace rs2; | |||
|
|
|||
| void BaseRealSenseNode::setup() | |||
| { | |||
| #if defined (ACCELERATE_WITH_GPU) | |||
| bool use_gpu_processing = (_accelerate_with_gpu == accelerate_with_gpu::GL_GPU); | |||
| initOpenGLProcessing(use_gpu_processing); | |||
There was a problem hiding this comment.
can we save this call if accelerate_with_gpu != GL_GPU ?
or, we must call it and init things there ?
There was a problem hiding this comment.
or, we must call it and init things there ?
We must call initOpenGLProcessing() and init things even for accelerate_with_gpu != GL_GPU case, because the filter objects are created from rs::gl::colorizer, rs2::gl::align and rs2::gl::pointcloud. classes
Arun-Prasad-V
force-pushed
the
gl_support
branch
from
7786a9a
to
73df46b
Compare
No description provided.