Calibration on Colab takes 30+ minutes instead of 15 as in docs
abitrolly opened this issue · 2 comments
I can also see this warning.
/usr/local/lib/python3.7/dist-packages/numba/np/ufunc/parallel.py:355: NumbaWarning: The TBB threading layer requires TBB version 2019.5 or later i.e., TBB_INTERFACE_VERSION >= 11005. Found TBB_INTERFACE_VERSION = 9107. The TBB threading layer is disabled.
warnings.warn(problem)
I will try to install this tbb and see if anything improves.
The notebook link is https://colab.research.google.com/drive/1N53E-qFSkOBcjB8IvXhBCmJPcisQmuxV?usp=sharing but its contents will most likely change.
Most of the time here is spent on detecting the charuco boards from the videos, which is primarily using OpenCV's library, independent of Numba, so I doubt that tbb will change much for the speed here.
We've started work recently on some better ChArUco board detectors using DeepLabCut which should be more robust and run faster, but it'll probably be a few months before we properly push them out.
How do you measure the performance of Numba? I want to bump it to the latest version, but it would be nice to see some graphs that it really speeds up anything. It am not sure I will be able to come up with the test notebook myself.