MrDeepFakes Forums
  • New and improved dark forum theme!
  • Guests can now comment on videos on the tube.
   
Fappenberg(outdated) DFL 1.0 SAE+ MOD.
#11
(11-29-2019, 02:58 PM)Gluke Wrote: You are not allowed to view links. Register or Login to view.
(11-07-2019, 04:27 AM)tutsmybarreh Wrote: You are not allowed to view links. Register or Login to view.I didn't noticed new update with the TrueFace strength setting. Aren't you iperov in disguise? Tongue
BTW since by default this is not in regular DFL (nor is an option to switch Adam/RMSProp) have you contacted iperov if he wouldn't want to add this by default to SAE and also bring TrueFace strength and likewise option to switch Adam/RMSProp to SAEHD? I know that he wants SAEHD to be the default/best model so adding this stuff to SAE might seem pointless but who knows. It's a cool option for people who don't want necessarily to retrain/make new SAEHD models and want to have some features of SAEHD in regular SAE.

Or is it actually pointless and you recommend users to just stick to SAEHD and treat this only as a "novelty" of sorts?

Also I sticked this thread so it always on top.

Based on my personal experience TrueFace is not so good if your faceset has many different sources couse it tends to get more inconsistent lighting in the results, maybe works better if you have just one video for src or the same lighting.
What're your thought?

Yes, upon using it more I too noticed True face can be a bit tricky to work with but I would go as far to say you needd perfect dataset, some color inconsistency can be handled by enabling rct, but with bad datasets which can't be improved, stuff with varying makeups/lighting conditions I'd probably not use TrueFace as it will make every angle shift between different looks (to a degree) and make it look weird.
Certainly it works best for frontal shots with uniform/frontal lighting, assuming same for source dataset.
Raising money for a new GPU, if you enjoy my fakes or my work on forums consider donating via bitcoin, tokens or paypal/patreon, any amount helps!
Paypal/Patreon: You are not allowed to view links. Register or Login to view.
Bitcoin: 1C3dq9zF2DhXKeu969EYmP9UTvHobKKNKF
Want to request a paid deepfake or have any questions reagarding the forums or deepfake creation using DeepFaceLab? Write me a message.
TMB-DF on the main website - You are not allowed to view links. Register or Login to view.
#12
(11-29-2019, 04:02 PM)tutsmybarreh Wrote: You are not allowed to view links. Register or Login to view.
(11-29-2019, 02:58 PM)Gluke Wrote: You are not allowed to view links. Register or Login to view.
(11-07-2019, 04:27 AM)tutsmybarreh Wrote: You are not allowed to view links. Register or Login to view.I didn't noticed new update with the TrueFace strength setting. Aren't you iperov in disguise? Tongue
BTW since by default this is not in regular DFL (nor is an option to switch Adam/RMSProp) have you contacted iperov if he wouldn't want to add this by default to SAE and also bring TrueFace strength and likewise option to switch Adam/RMSProp to SAEHD? I know that he wants SAEHD to be the default/best model so adding this stuff to SAE might seem pointless but who knows. It's a cool option for people who don't want necessarily to retrain/make new SAEHD models and want to have some features of SAEHD in regular SAE.

Or is it actually pointless and you recommend users to just stick to SAEHD and treat this only as a "novelty" of sorts?

Also I sticked this thread so it always on top.

Based on my personal experience TrueFace is not so good if your faceset has many different sources couse it tends to get more inconsistent lighting in the results, maybe works better if you have just one video for src or the same lighting.
What're your thought?

Yes, upon using it more I too noticed True face can be a bit tricky to work with but I would go as far to say you needd perfect dataset, some color inconsistency can be handled by enabling rct, but with bad datasets which can't be improved, stuff with varying makeups/lighting conditions I'd probably not use TrueFace as it will make every angle shift between different looks (to a degree) and make it look weird.
Certainly it works best for frontal shots with uniform/frontal lighting, assuming same for source dataset.

I do agree. When you mention "enabling rct" you mean during training or at the conversion stage?
#13
(11-29-2019, 08:41 PM)Yes, upon using it more I too noticed True face can be a bit tricky to work with but I would go as far to say you needd perfect dataset, some color inconsistency can be handled by enabling rct, but with bad datasets which can\t be improved, stuff with varying makeups/lighting conditions I'd probably not use TrueFace as it will make every angle shift between different looks (to a degree) and make it look weird. Wrote: You are not allowed to view links. Register or Login to view.Certainly it works best for frontal shots with uniform/frontal lighting, assuming same for source dataset.

I do agree. When you mention "enabling rct" you mean during training or at the conversion stage?

During training.
Raising money for a new GPU, if you enjoy my fakes or my work on forums consider donating via bitcoin, tokens or paypal/patreon, any amount helps!
Paypal/Patreon: You are not allowed to view links. Register or Login to view.
Bitcoin: 1C3dq9zF2DhXKeu969EYmP9UTvHobKKNKF
Want to request a paid deepfake or have any questions reagarding the forums or deepfake creation using DeepFaceLab? Write me a message.
TMB-DF on the main website - You are not allowed to view links. Register or Login to view.
#14
(11-29-2019, 09:57 PM)tutsmybarreh Wrote: You are not allowed to view links. Register or Login to view.
(11-29-2019, 08:41 PM)Yes, upon using it more I too noticed True face can be a bit tricky to work with but I would go as far to say you needd perfect dataset, some color inconsistency can be handled by enabling rct, but with bad datasets which can\t be improved, stuff with varying makeups/lighting conditions I Wrote: You are not allowed to view links. Register or Login to view.Certainly it works best for frontal shots with uniform/frontal lighting, assuming same for source dataset.

I do agree. When you mention "enabling rct" you mean during training or at the conversion stage?

During training.

Thank you, found out that it can change eyes and lips color. In my case it turn blue eyes to green...just wondering if it can be predicted as behaviour.
So I set to none, also in the conversion stage and use the istogram match masked.
#15
Hello, just made a small update that adds in the 'sot' color training option. Will work with the current version on GitHub. Should work on the versions between Nov 12-24, 2019.
#16
Hey I'm trying to figure all this stuff out and saehd and sae keep telling me it cant allocate 2.18gb of the card and giving me an error? I'm on a 2070 so I'm a bit confused as to what's going on
#17
Chers amis, j'ai formé un mannequin pendant 10 heures. Il a été sauvegardé automatiquement toutes les heures et j'ai exporté l'espace de travail à la fin.
Quand j'ai rechargé le modèle Après avoir installé model.py, tout est réinitialisé. J'ai perdu ma formation. As-tu une idée? Ce que je fais mal ? Je vous remercie ! Ps désolé pour l'anglais
#18
(12-02-2019, 06:40 PM)Gluke Wrote: You are not allowed to view links. Register or Login to view.
(11-29-2019, 09:57 PM)tutsmybarreh Wrote: You are not allowed to view links. Register or Login to view.
(11-29-2019, 08:41 PM)Yes, upon using it more I too noticed True face can be a bit tricky to work with but I would go as far to say you needd perfect dataset, some color inconsistency can be handled by enabling rct, but with bad datasets which can\t be improved, stuff with varying makeups/lighting conditions I Wrote: You are not allowed to view links. Register or Login to view.Certainly it works best for frontal shots with uniform/frontal lighting, assuming same for source dataset.

I do agree. When you mention "enabling rct" you mean during training or at the conversion stage?

During training.

Thank you, found out that it can change eyes and lips color. In my case it turn blue eyes to green...just wondering if it can be predicted as behaviour.
So I set to none, also in the conversion stage and use the istogram match masked.

The color changes is usually because the background of the facesets are not a neutral color. It sucks but you'll have to test the different conversion color options to try to fix it if you've already trained with RCT on.

(12-09-2019, 11:07 AM)hardcorenmore420 Wrote: You are not allowed to view links. Register or Login to view.Hey I'm trying to figure all this stuff out and saehd and sae keep telling me it cant allocate 2.18gb of the card and giving me an error? I'm on a 2070 so I'm a bit confused as to what's going on

Likely a OOM problem. Post your training settings for people to see. Lower your batch size.

(12-12-2019, 02:04 AM)FUNKYGUIZMO Wrote: You are not allowed to view links. Register or Login to view.Chers amis, j'ai formé un mannequin pendant 10 heures. Il a été sauvegardé automatiquement toutes les heures et j'ai exporté l'espace de travail à la fin.
Quand j'ai rechargé le modèle Après avoir installé model.py, tout est réinitialisé. J'ai perdu ma formation. As-tu une idée? Ce que je fais mal ? Je vous remercie ! Ps désolé pour l'anglais

Is this in google colab? If so it backs up to your google drive account.
~ Fake it till you make it ~
You are not allowed to view links. Register or Login to view.
You are not allowed to view links. Register or Login to view.
#19
Hi Friend , yes it' s in colab. Workspace is exporting to drive but when i import it more later, model is detected but training is reinitialized. Must i have to do something with model.py ? I must import it all thé Time i'm connect ? Save it After export in drive at the end . Thank you again
#20
I updated SAE+ with the changes iperov did with SAEHD and Quick96...

"improved model generalization, overall accuracy and sharpness
by using new 'Learning rate dropout' technique from paper You are not allowed to view links. Register or Login to view.
An example of a loss histogram where this function is enabled after the red arrow:
You are not allowed to view links. Register or Login to view."

Requires at least Dec 11, 2019 version of DeepFaceLab

Keep in mind the changes make the model run slower and uses more vram.

Forum Jump:

Users browsing this thread: 1 Guest(s)