Question on training.

If training is failing to start, and you are not receiving an error message telling you what to do, tell us about it here


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for reporting errors with the Training process. If you want to get tips, or better understand the Training process, then you should look in the Training Discussion forum.

Please mark any answers that fixed your problems so others can find the solutions.

Locked
User avatar
mhunt
Posts: 11
Joined: Mon Jun 28, 2021 5:09 pm
Been thanked: 1 time

Question on training.

Post by mhunt »

Hello, in the last month I was able to install FaceSwap, but using only the CPU, not my nvidia card since it failed to install with it due to a driver conflict.

For a start, I wanted to do something very simple like changing a single image, but the software would not work with single images, just videos, so I made a 5 second video of a single image for both source and destination.

So far training is up to 128k + but I don't see good results if I stop the training and do the conversion.

I used a simple tutorial from youtube. If any recommends changes in the setting for training, do I have to start all over ?

These are the settings I used:

EXTRACT

  • VideoA
    Extract faces from VideoA to ImgA

  • VideoB
    Extract faces from VideoB to ImgB

  • detector and aligner plugin set to S3Fd & Fan

  • masker set Unet-Dfl, VGG Clear & VGG-Obstructed

Start "Extract"

  • Generates: **A_alignments.fsa & **B_alignments.fsa

TRAIN

  • Faces/InputA <- ImgA (original)

    Face/Alignments A <- **A_alignments.fsa

  • Faces/InputB <- ImgB (replace A with B)

    Face/Alignments B <- **B_alignments.fsa

  • Model <- set to model Dir (ModelAB)

  • Trainer set to * original

  • Timelapse Input A <- ImgA

  • Timelapse Input B <- ImgB

  • Timelapse Output <- TimelapseFaceAB (make dir)

  • save project

Start "Train"


CONVERT

  • Data/ Input Dir <- VideoA

  • Data/ Output Dir <- Dest Dir

  • Data/ Model Dir <- ModelAB Dir

  • Color Adjustment

    Set AVG-Color (o Match-Hist)

  • Mask Type

    Set Extended

  • Writer

    Set FFmpeg


Start "Convert"

User avatar
torzdf
Posts: 2651
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 129 times
Been thanked: 622 times

Re: Question on training.

Post by torzdf »

You can't train Faceswap on a single image. The Neural Network needs lots of images at a high variety of data. Lots of identical images are not going to get it anywhere.

Related: viewtopic.php?f=27&t=831

My word is final

Locked