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.

Aloha

phoenix1969

DF Vagrant
I'm UFOlogist on yt
been playing with the software for a week, here's my first two video clips:
[video=youtube]


and this one was my 1st attempt, 


[video=youtube]

could use some advanced tips

these are on an EVGA lappy sporting a nvidia 1070 8g with an intel 6820hk cpu
the best settings i could get are: 
Model name: SAE

== Current iteration: 0
== Model options:
== |== batch_size : 5
== |== sort_by_yaw : True
== |== random_flip : False
== |== resolution : 256
== |== face_type : f
== |== learn_mask : True
== |== optimizer_mode : 1
== |== archi : df
== |== ae_dims : 256
== |== e_ch_dims : 21
== |== d_ch_dims : 10
== |== multiscale_decoder : True
== |== ca_weights : True
== |== pixel_loss : True
== |== face_style_power : 0.0
== |== bg_style_power : 0.0
== |== apply_random_ct : True
== |== clipgrad : True

and run for 10,000 iter then bump the face and bg style powers up to about 20 then ramp down every couple hundred iter, and settle at 1 on face and 0 on bg, all within 1000 iter, then just let simmer till 50,000
Any tips or comments appriciated   :cool:

CODE CONRIBUTIONS:  
5.3 data_dst sort by black
and
5.2.4 data sort by dissimilar face
yea i know, simple but they seem essential for sorting destination faces, yw.


next one a full length SNL cold-open, just beginning training on main character
 
Top