MrDeepFakes Forums

Some content may not be available to Guests. Consider registering an account to enjoy unrestricted access to guides, support and tools

  • We are looking for community members who are intested in helping out. See our HELP WANTED post.

Mini-models for 1050ti

VladShev

DF Vagrant
Verified Video Creator
Hi, I trained three small models for beginners. He specially trained on a weak video card, as evidence that even at 1050ti you can get at least some acceptable result. However, these models can also work on more powerful graphics cards. I hope they help beginners save a couple of days on pre-training and selection of settings.
Also, these models can be launched not only at 1050ti, but also on any other high-level graphics card.

Used PC:
 i5 8400 + 1050ti + 16gb ram

1) quick96_100k_iter:
3m7LhZ8h.jpg

Settings:
=============== Model Summary ================
== ==
== Model name: Quick96 ==
== ==
== Current iteration: 100062 ==
== ==
== ------------ Model Options ------------- ==
== ==
== batch_size: 4 ==
== ==
== -------------- Running On -------------- ==
== ==
== Device index: 0 ==
== Name: GeForce GTX 1050 Ti ==
== VRAM: 4.00GB ==
== ==
==============================================

Download link: https://mega.nz/#F!A5whCahR!QAqDzI4WKgELjY6EgTKKiw


_____________________
2) saehd-df-bs4_100k_iter:
dOwnyMWh.jpg

Settings:

=============== Model Summary ===============
==                                         ==
==         Model name: SAEHD               ==
==                                         ==
==  Current iteration: 100306              ==
==                                         ==
==------------- Model Options -------------==
==                                         ==
==        random_flip: True                ==
==         resolution: 128                 ==
==          face_type: f                   ==
==         learn_mask: False               ==
==     optimizer_mode: 2                   ==
==              archi: df                  ==
==            ae_dims: 256                 ==
==         ed_ch_dims: 21                  ==
==         lr_dropout: False               ==
==        random_warp: False               ==
== true_face_training: False               ==
==   face_style_power: 0.0                 ==
==     bg_style_power: 0.0                 ==
==            ct_mode: none                ==
==           clipgrad: True                ==
==         batch_size: 4                   ==
==                                         ==
==-------------- Running On ---------------==
==                                         ==
==       Device index: 0                   ==
==               Name: GeForce GTX 1050 Ti ==
==               VRAM: 4.00GB              ==
==                                         ==
=============================================

Download link: https://mega.nz/#F!plwTnCqZ!TuoEomGb_FsN6xpxU_B4ug


_______________________
3) saehd-liae-bs4_160k_iter:
I2bTVc2h.jpg

Settings:

=============== Model Summary ===============
==                                         ==
==         Model name: SAEHD               ==
==                                         ==
==  Current iteration: 162841              ==
==                                         ==
==------------- Model Options -------------==
==                                         ==
==        random_flip: True                ==
==         resolution: 128                 ==
==          face_type: f                   ==
==         learn_mask: False               ==
==     optimizer_mode: 2                   ==
==              archi: liae                ==
==            ae_dims: 256                 ==
==         ed_ch_dims: 21                  ==
==         lr_dropout: False               ==
==        random_warp: False               ==
== true_face_training: False               ==
==   face_style_power: 0.0                 ==
==     bg_style_power: 0.0                 ==
==            ct_mode: none                ==
==           clipgrad: True                ==
==         batch_size: 4                   ==
==                                         ==
==-------------- Running On ---------------==
==                                         ==
==       Device index: 0                   ==
==               Name: GeForce GTX 1050 Ti ==
==               VRAM: 4.00GB              ==
==                                         ==
=============================================

Download link: https://mega.nz/#F!YlBVQAKa!UHV4U4ojXO_BkbjgTmzwiw
 

VladShev

DF Vagrant
Verified Video Creator
Джиги said:
попробуйте достичь 100k выше для итерации .. это будет выглядеть лучше

Yes, I’m now retraining models up to 100k in connection with the latest update deepfacelab from 12/28/2019, because after optimization they need to be retrained from the beginning. I will soon post updated models.
 

VladShev

DF Vagrant
Verified Video Creator
Hi, all previous models are very outdated (they were made for dfl 1.0 version).

Now, after 8 months, a lot has changed. The developer has improved the speed and quality of training, and even added the ability to use "-d" models, which allows you to train on 4 gigabyte cards not 128 (df) resolution, but 256 (df-ud). For video cards with low video memory, this is very cool.
I am currently training such a model for a million iterations, and after I test it in work, I will post it here.
liae-ud can also be done, but probably in a lower resolution, due to the heavier architecture
 

telorjin

DF Vagrant
VladShev said:
Hi, all previous models are very outdated (they were made for dfl 1.0 version).

Now, after 8 months, a lot has changed. The developer has improved the speed and quality of training, and even added the ability to use "-d" models, which allows you to train on 4 gigabyte cards not 128 (df) resolution, but 256 (df-ud). For video cards with low video memory, this is very cool.
I am currently training such a model for a million iterations, and after I test it in work, I will post it here.
liae-ud can also be done, but probably in a lower resolution, due to the heavier architecture

ok.. im gonna love it dude
 

VladShev

DF Vagrant
Verified Video Creator
Here's a million iteration model.
Trained at 8 batch size for greater accuracy. For use on 4GB graphics cards, just drop the batch size down to 4.

https://mega.nz/folder/hgJyXAoR#6bomF2dplFTwOVbk5iaZ1w

================= Model Summary =================
==                                             ==
==            Model name: 256-df-ud_SAEHD      ==
==                                             ==
==     Current iteration: 1028407              ==
==                                             ==
==--------------- Model Options ---------------==
==                                             ==
==            resolution: 256                  ==
==             face_type: f                    ==
==     models_opt_on_gpu: True                 ==
==                 archi: df-ud                ==
==               ae_dims: 256                  ==
==                e_dims: 64                   ==
==                d_dims: 64                   ==
==           d_mask_dims: 22                   ==
==       masked_training: True                 ==
==             eyes_prio: False                ==
==           uniform_yaw: True                 ==
==            lr_dropout: n                    ==
==           random_warp: False                ==
==             gan_power: 0.0                  ==
==       true_face_power: 0.0                  ==
==      face_style_power: 0.0                  ==
==        bg_style_power: 0.0                  ==
==               ct_mode: none                 ==
==              clipgrad: False                ==
==              pretrain: True                 ==
==       autobackup_hour: 0                    ==
== write_preview_history: True                 ==
==           target_iter: 0                    ==
==           random_flip: True                 ==
==            batch_size: 8                    ==
==                                             ==
==---------------- Running On -----------------==
==                                             ==
==          Device index: 0                    ==
==                  Name: Tesla P100-PCIE-16GB ==
==                  VRAM: 15.90GB              ==
==                                             ==
=================================================
 

skrillex

DF Vagrant
Hi, all previous models are very outdated (they were made for dfl 1.0 version).

Now, after 8 months, a lot has changed. The developer has improved the speed and quality of training, and even added the ability to use "-d" models, which allows you to train on 4 gigabyte cards not 128 (df) resolution, but 256 (df-ud). For video cards with low video memory, this is very cool.
I am currently training such a model for a million iterations, and after I test it in work, I will post it here.
liae-ud can also be done, but probably in a lower resolution, due to the heavier architecture
edit: nvm
 
Top