Hi everyone! Sorry for the wait, I'm looking into the changes now. It looks like upstream removed support for bodypix, and moved to a different library - I'm gonna see if there is still a benefit to using the bodypix version with GPU, as the new library doesn't seem to have GPU acceleration.
If there is reason to use both versions, then I split off this package into one that uses latest master, and another which supports the "legacy" bodypix version.
For anyone else having issues with bodypix dying with error:
Will fix if this issue is still present.
v4l2loopback should be optional, akvcam is a complete v4l2loopback fork and replacement.
Will fix after the first issue is addressed.
This needs
python-pip
to build.
Ditto.
but how are we meant to use the /opt dependencies instead?
The way it's meant to work is that I made a small Python launcher script that should prioritize those dependencies. I'll investigate this when I can.
Thanks for the feedback all!
Pinned Comments
CodingKoopa commented on 2021-02-16 04:13 (UTC) (edited on 2021-06-24 01:42 (UTC) by CodingKoopa)
THIS PACKAGE DOESN'T CURRENTLY WORK - NEEDS TO BE UPDATED FOR UPSTREAM CHANGES
GENERAL USAGE INFORMATION:
bodypix
andfakecam
binaries.USING GPU ACCELERATION (NVIDIA ONLY):
Tensorflow requires an older version of CUDA, 10.0, and, by extension (due to package requirements), cuDNN.
gcc7
AUR package. This is required for the older CUDA package to work.cuda-10.0.130-2-x86_64.pkg.tar.xz
,cudnn-7.4.2.24-1-x86_64.pkg
. You can automate the process of downloading and installing them like so:/etc/pacman.conf
to prevent the packages from being upgraded: