Package Details: baballonia v1.1.1.0rc6-4

Git Clone URL: https://aur.archlinux.org/baballonia.git (read-only, click to copy)
Package Base: baballonia
Description: A cross-platform, hardware-agnostic VR eye and face tracking application.
Upstream URL: https://github.com/Project-Babble/Baballonia
Keywords: babble eye face tracking vr
Licenses: LicenseRef-Babble Software Distribution License 1.0
Submitter: awh
Maintainer: awh
Last Packager: awh
Votes: 3
Popularity: 1.15
First Submitted: 2025-10-29 02:56 (UTC)
Last Updated: 2026-03-23 22:02 (UTC)

Latest Comments

1 2 3 Next › Last »

awh commented on 2026-03-23 22:09 (UTC) (edited on 2026-03-23 22:10 (UTC) by awh)

The package is now split up into baballonia, babbletrainer, and babblecalibration-bin (which provides babblecalibration). If someone wants to make and maintain a non -bin version of babblecalibration, go ahead. I left my starting attempt in the comments there, but it's pretty broken.

As for this new split package, I was able to run the app, but didn't test it worked past that.

awh commented on 2026-03-23 18:19 (UTC)

The package is currently building the 1.3.8-linux-paths tag for BabbleTrainer, and BabbleCalibration is just unpacking the GitHub release zip for now since I didn't have time to look into how to build it myself. After that, I was planning to separate BabbleTrainer and BabbleCalibration into their own packages, with this one depending on them.

I'll message you regarding CI automation, it sounds like a good idea to me.

dfgHiatus commented on 2026-03-23 17:37 (UTC)

Hello, dfgHiatus from the Baballonia GitHub. The current trainer repo is a mess, I apologize - your best bet is to use the 1.3.8 release, we're looking to replace this entirely with something faster and leaner in the next month or two.

I've seen the comments below about GPU acceleration on Linux. I've compiled two draft PR with my findings here: - https://github.com/Project-Babble/Baballonia/pull/247 - https://github.com/Project-Babble/Baballonia/pull/249

Regarding "Input shape does not match initial shape. Expected: 6, got: 2", this can be resolved by deleting the LocalSettings.json file under the .config/ProjectBabble/ApplicationData folder, models trained from beta builds cannot be used with the production app.

Feel free to contact me, it would be nice to integrate some GitHub actions to automatically deploy updates when we push things. If we could add aarch64 to this as well that would be excellent - we support this and Steam Frame is just around the corner.

exuvo commented on 2026-03-18 21:26 (UTC)

With v1.1.1.0rc6 the wrong trainer seems to be used. When loading a newly trained model i get "Input shape does not match initial shape. Expected: 6, got: 2". I think they went back to the veryold branch as they felt the new one was not ready yet. If i manually train on the same training data with veryold-branch trainer (patched to opset 18 as he skipped that in the merge) i get one with 6 shapes that loads up without errors.

Trying atm to figure out with upstream what the intended version mix is as there is no release for that trainer what i can see?, i will get back to you.

awh commented on 2026-01-31 23:01 (UTC)

@hype-vhs added - not sure why namcap didn't catch that.

hype-vhs commented on 2026-01-31 22:48 (UTC)

It fails to build in a clean chroot, please add unzip to makedepends?

exuvo commented on 2025-12-22 18:36 (UTC)

I built the new package and it works. Ran the trainer manually from terminal with my previous training data and everything ran as expected on GPU. Thanks for your hard work! Have a good christmas!

awh commented on 2025-12-22 18:14 (UTC) (edited on 2025-12-22 18:15 (UTC) by awh)

I released v1.1.0.9rc3-3 - let me know if it works for you. The solution actually ends up being quite a bit nicer, since the GPU support can be swapped in via alternate providing Arch packages, and we don't have to build a giant blob in this package.

I might look at building BabbleCalibration from source at some point too, but for now it works.

exuvo commented on 2025-12-21 16:53 (UTC)

The one based on https://github.com/Project-Babble/BabbleTrainer/blob/main/requirements_gpu.txt with CUDA only support i think? was too large to build on github the dev said. The version under releases is still only CPU for non-windows.

As far as i understand pytorch has no version that supports both NVIDIA and AMD at the same time for non-windows. On windows it uses a microsoft directml layer between to support multiple GPU vendors.

awh commented on 2025-12-20 19:42 (UTC) (edited on 2025-12-20 19:43 (UTC) by awh)

I pushed a pkgrel update to fetch the prerelease version of BabbleTrainer. If the prebuilt binary doesn't include the necessary libraries for Linux GPU support then that should be fixed upstream, assuming there's no platform specific requirements that means it can't be built generically.

edit: or do you mean if those Arch packages are installed, the prebuilt version works fine?

1 2 3 Next › Last »