Impact of workitem_size_cameras, workitem_size_pairs, max_workgroup_size

[SFL_Beda]

Does somebody know the impact of the variables
workitem_size_cameras
workitem_size_pairs
max_workgroup_size

which are used in many functions of the python api?

Trough trial and error I have found out that workitem_size_cameras impacts how many images are processed in a package on the gpu. Afterwards it does initialize the gpu again "Found 1 GPUs in 0.000307 sec (CUDA: 0.000187 sec, OpenCL: 0.000107 sec)".

Now the interesting thing for me is that the defaults in the GUI don't seem to correspond to the defaults in the python module.
If I process a project in the gui, it extracts keypoints from all pictures in one go while in the python module it does it in sets of 20.

now my question is: How should these variables be used? does the GUI calculate ideal values based on the available hardware?

[rhseaeng]

I am wondering the same thing. It seems that my network processing is substantially slower via Python than via the GUI, without specifying these parameters in Python.

[SFL_Beda]

Did you find differences when adjusting these?

In my very limited testing it didn't seem to really make a difference.