Code segment for ICP improvement? about semantic_suma HOT 4 CLOSED

Hey @anthonypan08,

Thanks for using our code. We actually use OpenGL to accelerate the calculation of the jacobians in Frame2Model_jacobians.geom. For the entire frame-to-model ICP process, you may also have a look at Frame2Model.cpp, LieGaussNewton.cpp and also the updatePose() function in SurfelMapping.cpp.

Hope it helps!

from semantic_suma.

It seem like Frame2Model.cpp utilizes the Frame2Model.geom file and the weights of each point are kept track in Frame2Model.geom . From what I can see, the .xyz symbolizes the 3D coordinate of each point in point cloud and the .w is the weights. Also can you tell me where those properties are defined (I am quite new to OpenGL so forgive me if the questions are elementary)

My plan is to use expectation maximization algorithm to soft assign the weight of each point. Is it correct that I would have to modify the contents in the OpenGL Frame2Model.geom? I want to ensure that each point holds an array of possible weights instead of just 1 weight. Do know any specific issue that would arise if I just go ahead and change the data structure in OpenGL?

from semantic_suma.

Yes, you are right. You could find the definition of Surfel properties in Surfel.h.

Maintaining an array for each point is not that straightforward, but it is definitely doable. You could find more information here

from semantic_suma.

Since there is no update, I'm going to close this issue. Please feel free to ask me to reopen it if needed.

from semantic_suma.