Mr DeepFakes Forums

Guests - Sign up to view links and videos on the forums. Please check your spam mail. (February 26, 2018)

  • New and improved dark forum theme!
  • Guests can now comment on videos on the tube.
   
  •  Previous
  • 1
  • 2
  • 3
  • 4(current)
  • 5
  • 6
  • ...
  • 139
  • Next 

Shoutbox archive

Go to last read | Mark all read

avatar
the increase resolution will still be there, just extract in png
avatar
put your upscalled images into /data_src - re-extract, it will be place in /data_src/aligned
avatar
Wait. I'm confused. I'll explain my process...
avatar
well, it's not all lossed, you can extract those upscaled
avatar
That's the part that just bit me in the ass...lol
avatar
To save a massive amount of time, I upscaled the alligned faces, not the source images.
avatar
its okay, its a learning process
avatar
just take the new upscaled and extract them again
avatar
Well rip, I modified all my source images, upscaled and cleaned...2 days of work down the drain, as the images error on load. FML...rofl
avatar
Ahh...Nvm. I just saw the new one does the same as well.
avatar
and the older method, did it use to save them externally on another image with guidelines?
avatar
@Kamrooz yes
avatar
Does the S3FD method save the mask internally for the images when you extract face?
avatar
and I always start with it off while styling powers are on, then switch if on once they start glitching
avatar
Haven't had a single crash with pixel loss.. using 256 though.
avatar
@ALL !!!!!! PREY FOR NOTRE DAME !!!!!!
avatar
@titan_rw 3 times BS 6 equals 2400. 1 times BS 18 equals 2400. Mathematically, and theorically it is the same. However, according to @dpfks guide, larger batch sizes yield better results. So if there's not difference in speed, I'd go with the larger BS. Keep in mind that I don't use DFL, so free advice is worth what you pay for it.
avatar
@titan_rw theoretically it's the same.
avatar
@Cerciusx there's a bit of jitter with the example you posted. Was that seamless convert?
avatar
What do you guys think is faster, BS 6 taking 800ms, or BS 18 taking 2400 ms? I know the lower the batch size the more iterations it's going to take, but when iterations are 3x as fast, which one would be 'faster'?

  •  Previous
  • 1
  • 2
  • 3
  • 4(current)
  • 5
  • 6
  • ...
  • 139
  • Next