Windows app fails to launch correctly (RTX 30 series graphics cards)

[kaiwildberger]

I've installed the windows app, and when I launch it either from the shortcut or the executable, no window pops up.
I can see it in task manager, but nowhere else, and I left it for ~15 minutes and it stayed like this.

I have a 3060Ti with 8gb of ram on Windows 10.

[alievk]

Did you try to kill and restart it?

[kaiwildberger]

Yes, I tried restarting avatarify and my computer as well. I also reinstalled to a different location, but none of that made any perceivable difference.

[Justan-alt]

For me on windows 10, a window with my webcam feed pops up and it doesnt respond to anything. I left it for 10 min But it still didnt respond. I tried killing it in task manager, reinstalling everything but still didnt work. And I have a rtx 3070

[alievk]

Thanks for the feedback, we will take this into account. Unfortunately, there's nothing else I could recommend at the moment.

[sjhill01]

Same here. I'm on a brand new PC with an i9-10900 and an RTX 3070.

If you leave it long enough, the face calibration window comes up, and that works ok, but then it fails to detect a face and eventually hits a runtime error (there's a several minute delay between the "no face detected" and the error:

(avatarify) C:\Users\Steve\avatarify>run_windows.bat
[1609855695.672446] Loading Predictor
C:\Users\Steve\avatarify\afy\predictor_local.py:34: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details.
config = yaml.load(f)

=== Control keys ===
1-9: Change avatar
1: avatars\einstein.jpg
2: avatars\eminem.jpg
3: avatars\jobs.jpg
4: avatars\mona.jpg
5: avatars\obama.jpg
6: avatars\potter.jpg
7: avatars\ronaldo.png
8: avatars\schwarzenegger.png
W: Zoom camera in
S: Zoom camera out
A: Previous avatar in folder
D: Next avatar in folder
Q: Get random avatar
X: Calibrate face pose
I: Show FPS
ESC: Quit

Full key list: https://github.com/alievk/avatarify#controls

Warning: No faces were detected.
Traceback (most recent call last):
File "afy/cam_fomm.py", line 312, in
out = predictor.predict(frame)
File "C:\Users\Steve\avatarify\afy\predictor_local.py", line 82, in predict
use_relative_jacobian=self.relative, adapt_movement_scale=self.adapt_movement_scale)
File "C:\Users\Steve\avatarify\fomm\animate.py", line 33, in normalize_kp
jacobian_diff = torch.matmul(kp_driving['jacobian'], torch.inverse(kp_driving_initial['jacobian']))
RuntimeError: inverse: For batch 0: U(1022296064,1022296064) is zero, singular U.

[kaiwildberger]

Could it be pytorch and compute compatibility 8.6 on the 30 series? I left it for 15+ minutes and had a similar experience to what sjhill01 described.

[matts5323]

The same problem, tried installing the windows app multiple times, it shows in task manager it is running, and my webcam is has been turned on, but there is no visible app on the screen. In on a brand new Pc with a 3060ti and a Ryzen 5 3600.

[jlaframboise]

I have the same problem. I also tried on Ubuntu 20.04 and had the same issue where it stalled forever at the yaml load.
(5600X and rtx 3070)

[ghost]

SAME. Stalled at yaml.load as well followed by fails to detect a face/cam not responding i910900k rtx 3090

[ochen1]

I'm pretty sure it is stalling at model download rather than yaml load. Try waiting longer perhaps?

…

On Tue, Jan 12, 2021, 3:17 PM chapoogie, ***@***.***> wrote: SAME. Stalled at yaml.load as well followed by fails to detect a face/cam not responding i910900k rtx 3090 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#335 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AOHGCDMHSYYGK4G5F3MZVKDSZTC6DANCNFSM4VN3BWLQ> .

-- If you believe this email violates Canada's Anti-Spam Legislation (CASL), please contact us <https://docs.google.com/forms/d/e/1FAIpQLSemiJbGqN-4ClXBDKPoRp4-EKTswV5sSzp-KE6ff9LW64IMiA/viewform?usp=sf_link>. Edmonton Public Schools, Centre for Education, 1 Kingsway NW, Edmonton AB T5H 4G9

[JohanAR]

Updating title since everybody reporting this seems to be using a RTX 30 series GPU

[JohanAR]

@sjhill01 @jlaframboise it looks like you're running the Python version of Avatarify, while the creator of this issue is using the Window app version of Avatarify. The good news is that the Python version has been updated, with a potential fix for RTX 30 series GPU owners. Please pull latest master and try again if you're using the Python version.

[jlaframboise]

@JohanAR Thank-you it works great now! A note on performance is it seems to be relatively CPU bottlenecked but this is a really cool piece of software, kudos!

[zarklon]

Sorry to be a complete noob here when it comes to this.. but back when I had a 2060 card and followed the tutorial closely to install the python version it worked.. Now I have a 3070 and it doesnt. How does one "Pull the latest master" ? Is this done via a command in the Miniconda thing that I normally type run_windows.bat in to get things going? If so , what is that command that will "Pull the master" and update things for 3070 cards? Is this an easy thing to explain to a complete noob or should I just follow all the steps in the tutorial again and that will be like doing what you're saying? Any help would be appreciated as the desktop version of avatarify just doesnt run for me.

My screen is just stuck at

[1642671149.055903] Loading Predictor
C:\Users\Zarkl\avatarify\afy\predictor_local.py:34: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details.
config = yaml.load(f)

and does nothing
Windows 10 - RTX 3070 TI here

[JohanAR]

@zarklon if you have used the git program to clone the repository, then you run "git pull" to update your local copy with the latest changes on the server. However if you've download avatarify-python as a zip file I think you need to re-download the whole thing to get the latest version

[zarklon]

Ok I ended up actually figuring that out with a few google searches last night and I got it to update but then it took like 30 minutes to come up for me and failed to face track. Left me with this error when it failed to calibrate .. And idea what to do at this point? And why in the world did it take forever to launch? Should I go and delete the whole directory and just follow the tutorial again? I got so far.. and tried so hard..

[JohanAR]

I think the first time you run the script it downloads pre-trained data for the ML algorithms, which could explain why it took so long time. I think the server that hosts the file occasionally has some SSL certificate errors too.

Unfortunately this is not an error I recognize. Perhaps try doing a clean install, and if it still doesn't work open a new issue on GitHub.