RigidCPD does not work in multiprocessing.Process

[jwindhager]

When running RigidCPD in a multiprocessing.Process (Linux, start_method="fork") without CUDA/cupy available, the process hangs in the following line:

probreg/probreg/math_utils.py

Line 26 in 21d5ed2

return _math.squared_kernel(x, y).sum() / (x.shape[0] * x.shape[1] * y.shape[0])

Debugging & inspecting the stack trace, this gets called here:

probreg/probreg/cpd.py

Line 130 in 21d5ed2

sigma2 = self._squared_kernel_sum(self._source, target)

Presumably, the C code does not work with Python's multiprocessing module?

[neka-nat]

Hi,
The problem is that the c wrapper object cannot be pickle when using multiprocessing.
I don't know if it will work, but the following articles and tools may help.

https://stackoverflow.com/questions/19984152/what-can-multiprocessing-and-dill-do-together
https://pathos.readthedocs.io/en/latest/pathos.html
https://joblib.readthedocs.io/en/latest/auto_examples/serialization_and_wrappers.html

[jwindhager]

Thanks a lot for the insight! I worked around the problem by spawning the worker processes instead of forking them.