victorprad / InfiniTAM

A Framework for the Volumetric Integration of Depth Images
http://www.infinitam.org
Other
918 stars 351 forks source link

Problems about learning the code #51

Closed MarkZurkberg closed 7 years ago

MarkZurkberg commented 8 years ago

I have compiled InfiniTAM-spaint_v2, and I am trying to understand the code nowadays. Due to so many complex source files I found it's difficult to learn, although I have generate doxygen files it seems not useful for my learning process. Can somebody tell me how to learn the code and master the core algorithm? Thanks!

sgolodetz commented 8 years ago

The main pipeline is in ITMBasicEngine<...>::ProcessFrame. I would start from there and drill down, trying to figure out how things correspond to the descriptions in the various papers as I go.

An understanding of the way in which the CPU/GPU code is structured is also probably useful - basically you have an interface class, CPU and CUDA classes that derive from it, and shared code that is called by the implementations of the various member functions in the CPU and CUDA classes.

MarkZurkberg commented 8 years ago

I have used glm to load a 3D object successfully, but I don't know how to integrate this with InfiniTAM so now it is just limited in its own window. I want to find the entry point for the reconstruction scene so that I can add some furnitures into InfiniTAM. Could you please give me some advice? Thanks! @sgolodetz

sgolodetz commented 8 years ago

Depending on the format of the object, this may be more or less tricky - but in terms of adding things to the scene, the best places to look in the code are probably the bits to do with the scene representation (drill down from ITMScene) and the scene reconstruction engine (drill down from ITMSceneReconstructionEngine).