Faceswap Forum

just upgraded from Ubuntu 18 >19 > 20.04 LTS

Faceswap survived transitioning to 19.
Not to 20.04.
just reinstall and all is well.

I have trained for 20K iterations, and recovered images are quite better, the loss for each is about 0.039. I want to it would got better if I train longer, and when to stop ? Depends on loss? Although it may look good before a million :-) I've done it couple around 80k, with excellent facial data,...

Grassone wrote: ↑

Fri May 01, 2020 12:33 pm

P.S.
Any update on the new manual alignment tool? How is it going ?

Become a patreon and find out. Buy these fellas some coffee.

I tried several times.

I see many people asking about how long to train, iterations, batch sizes ect. Thought I would share to give people an idea of results. For this I was testing DFL-SAE. This was input size 192, and batch of 6. Roughly 7K images on both for various sources. I updated the extracted faces and alignments...

When I run the preview. Results show blank. Very clean XUbuntu and Faceswap install. works fine in windows 10. http://imgur.com/r3WTxanl.png This is the Facebook Log. 04/22/2020 20:12:20 MainProcess MainThread logger log_setup INFO Log level set to: INFO 04/22/2020 20:12:21 MainProcess MainThread fs...

I've been at this a couple months now. Enjoying the technology and getting a handle on how it works. Made a couple really good swaps @ 70K iterations. Made several other models just trying to teach myself. but I get a little dismayed.. Still blurry or somehow it is not right. So I set up a set of re...

Also got to figure out why time lapse doesn't want to work for me odd little hiccup there.

Actually doing that myself just to get a feel for how different models and mask work.
Very time-intensive but since I work 12-hour shifts I can at least get the training done .
Mainly just original, dfl-sae , dlight, and unbalanced. 8gb only goes so far

I think I know what you did. so you train your model. Then you want to apply that model to a video... You (currently)have to extract the alignments for every frame in the video. Extract Every N = 1. That generated alignments file is the one you need to give to the convert process. Then you need to c...

So, what I have been doing is I make a big folder I call "original". In there I put everything: google images, Video extracts, etc. Extract it all at 512pix then clean that "originals" folder all up. I do not separate the faces except for beards, glasses, or heavy makeup. So then I can train with on...

Oh it would be text with image examples. but for the most part the existing guide is quite informative, just easy to overlook/misinterpret one little detail and you've made a mass. That's reading comprehension with your not exactly familiar. I didn't 100% "get" alignments 100% for a bout a week . Ke...

I've been watching this board and the discord daily for the last few weeks while I'm learning all these concepts myself. The guides are excellent. I've read through them several times , especially when i'm stuck. I have caught up pretty well, although I was in commercial hardware solutions for 15 ye...

Model, settings?

Made this for my coworkers birthday. It really looks like her. Uncanny in some frames. https://www.youtube.com/watch?v=CNHrmynFT9c https://www.youtube.com/watch?v=CNHrmynFT9c DFL-SAE : tweaked some settings to run a lil better with my limited hardware. 65K iterations 128 x 128 samples. about 3000 fo...

03/24/2020 19:12:13 MainProcess _training_0 _base name DEBUG model name: 'dfl_sae' 03/24/2020 19:12:13 MainProcess _training_0 _base add_network DEBUG name: 'decoder_a', filename: 'dfl_sae_decoder_A.h5' 03/24/2020 19:12:13 MainProcess _training_0 _base __init__ DEBUG Initializing NNMeta: (filename:...

OK. did some rearranging, received a log and error! Both P106-90 show up in device manager. Working correctly. Part of Crash log 03/24/2020 18:59:57 MainProcess _training_0 _base name DEBUG model name: 'dfl_sae' 03/24/2020 18:59:57 MainProcess _training_0 _base add_network DEBUG name: 'decoder_b', f...

I did reduce the power needs. used the benchmark to ramp things up first. Pretty sure there is no 'brownout' hurtle. Everything is closed so I have had time today to fool around. I did notice the GPU are numbered: #0 p106 #1 Rx570 #2 p106 Now I am running FaceSwap on GPU 0, folding@home on GPU 1 & 2...

Because the cards have no display I used "geekbench" to benchmark the GPU for CUDA. Runs fine on the second card while the first card is running faceswap. (Also gets HOT). Even overclocked both about 2% to see if I could cause a failure. I can clearly run both. Is their a way to manually configure w...

1000 Watt I've reverted all the drivers.
Ill give it a shot with the benchmark Idea. Thank you.