Search found 21 matches

by tochan
Fri Oct 21, 2022 7:01 am
Forum: Swap Showcase
Topic: Second swap - Looking for advice (Kobe Bryant x Michael Jordan)
Replies: 1
Views: 21537

Re: Second swap - Looking for advice (Kobe Bryant x Michael Jordan)

If there are any tips to improve high frequency detail, I'm super down to hear it ! Hi, test this: Configure Settinings, Train, Global, Loss ==> take Loss Function "3" = Lpips_alex. Don't use hi nummers. Start with 2 (i use 1). High creates artetfacts Batch size go down form 16 to 12 with...
by tochan
Sun Sep 04, 2022 8:52 am
Forum: Training Discussion
Topic: [Discussion] Notes on Loss functions
Replies: 13
Views: 28429

Re: Notes on Loss functions

Thanks for this. I think a discussion on Loss Functions is well worthwhile. I have had some success with the following combination: Many thanks for this information. I train my disney model (512) 964.000 times in default settings and i miss the details (eyelashes etc.). with your example, there det...
by tochan
Wed Aug 31, 2022 6:47 am
Forum: Training Discussion
Topic: [Guide] Introducing - Phaze-A
Replies: 89
Views: 165008

Re: [Guide] Introducing - Phaze-A

Hi folks, after a long time i come back to faceswap ;). I tested the phaze-A "disney 512 model " in default settings from the config file. Here some experinces: Sys intel 12700KS, 64GB Ram, Nvidia 3090 FE. Batchsize 16 works fine (36-40 Egs/sec) after 575.000 iterations, the model run in a...
by tochan
Wed Apr 29, 2020 12:37 pm
Forum: Training Discussion
Topic: Yikes! The Spikes"
Replies: 4
Views: 3479

Re: Yikes! The Spikes"

same here. one time, i see this "live" in the preview image. at this moment / the next preview pic shows wrong colors hat the faces. don't know why

by tochan
Sat Mar 07, 2020 8:46 am
Forum: Training Discussion
Topic: What are the efficiency of two graphics cards?
Replies: 1
Views: 2807

Re: What are the efficiency of two graphics cards?

here are some results with my 1080, 2080, 1080+2080 and 2x2080 setup. https://forum.faceswap.dev/viewtopic.php?f=16&t=130 For training and convert: 2 cards are much faster than one 2080ti i think, but for extract you can use only one GPU. For training you can double the batch size with 2 cards(2...
by tochan
Sat Feb 22, 2020 10:00 am
Forum: Extract Discussion
Topic: It's possible to change face landmarks after extract?
Replies: 1
Views: 3192

It's possible to change face landmarks after extract?

Hi,

is it possible to change / optimize the Landmark possiton after the outo- extraction pic by pic?
For a check it's possible to write the Points in the extractet Pic (Debug Landmarks) but can i fix it when is found a mistake?

by tochan
Sat Jan 25, 2020 8:49 am
Forum: Training Support
Topic: CuDNN error when using 2 GPUs
Replies: 1
Views: 1917

CuDNN error when using 2 GPUs

Hi, i want to start a new Dlight try but it crash when i use 2 GPU's... (2x2080 RTX) One GPU works...Some Ideas? 2020-01-24 23:16:13.477575: E tensorflow/stream_executor/cuda/cuda_dnn.cc:82] CUDNN_STATUS_EXECUTION_FAILED in tensorflow/stream_executor/cuda/cuda_dnn.cc(3765): 'cudnnPoolingForward( cud...
by tochan
Tue Jan 21, 2020 8:06 pm
Forum: General Discussion
Topic: Artificially create more source images for Training.
Replies: 1
Views: 3455

Re: Artificially create more source images for Training.

At the moment, only on iOS... but... and now Android "now Early Access" i see!!! Now i can test the app on Andoid. Thanks for the reminder ;)

by tochan
Mon Jan 20, 2020 10:05 am
Forum: General Discussion
Topic: Idea to creat more source Images for Training
Replies: 0
Views: 5092

Idea to creat more source Images for Training

Hi, i use a Idea to create missing angles for the face training. The 3D Face Reconstruction tool know everyone i think. It works fine but for a faceswap with few images from the person, it's difficult to create a good swap. My idea is (never se this befor), take some fotos from other persons for the...
by tochan
Sun Jan 12, 2020 5:44 pm
Forum: Convert Discussion
Topic: Unet-Dfl Mask Type in Convert
Replies: 3
Views: 5052

Re: Unet-Dfl Mask Type in Convert

Thanks for implement the Unet-Dfl mask. Results looks very good :)

by tochan
Sat Jan 11, 2020 3:58 pm
Forum: Extract Discussion
Topic: Using multiple videos (upd)
Replies: 7
Views: 8680

Re: Using multiple videos (upd)

Extract the two videos (important: choose different file names for the videos !!! (Vid_1 / Vid_2, for example) Rename the two alignmets.fsa files (to alignments_vid_1.fsa / alignments_vid_2.fsa, as an example) Now copy the extracted Faces into one folder (when you use same video filenames, you'll o...
by tochan
Fri Dec 13, 2019 7:10 pm
Forum: Convert Discussion
Topic: Unet-Dfl Mask Type in Convert
Replies: 3
Views: 5052

Unet-Dfl Mask Type in Convert

Hi! first sorry for my bad english.... It's possible to use the Unet-Dfl masker form the extract in the convert job as an "layer"? The Unet-dfl hide fingers etc. in the preview window very well when i train the model, so my idea is to usw this "orignal areas they hide" (fingers e...
by tochan
Fri Nov 22, 2019 8:50 am
Forum: Hardware
Topic: GPU Training Speed
Replies: 11
Views: 19718

Re: GPU Training Speed

Hi again, little info form my side with some "new" parts in my System. Switch the AMD 1800x to 3800x Change the GPU from 2080+1080 to SLI 2080. Training works fine with SLI option is on in the OS with NVlink birge but there is no benefit for the training speed with ore without NVlink. Here...
by tochan
Sun Nov 17, 2019 8:54 pm
Forum: Hardware
Topic: GPU Training Speed
Replies: 11
Views: 19718

Re: GPU Training Speed

little update after my 1080 return form the repair. At the moment, i train a Dlight model with this 2 cards (now it works) general Dlight model, allow Growth "on" Oprtimizer Savings. Batch Size 39 Gpus 2 Train Plugin Global Coverage "87.5" Masktyp "compents" Subpixel Up...
by tochan
Wed Nov 13, 2019 7:37 pm
Forum: Convert Support
Topic: OOM out of memory during convert but not training
Replies: 14
Views: 17671

Re: Dlight Convert Crash: report

I try a reboot, same crash. I try a snapshot, same crash. Then i set the trainer plugin to default, create a new model an it works! Then i try the settings step by step with new Dlight models and the crash come, when the "Reflect Padding" is activ. Not testet with other trainers to this mo...
by tochan
Sun Nov 10, 2019 7:59 am
Forum: Convert Support
Topic: OOM out of memory during convert but not training
Replies: 14
Views: 17671

OOM out of memory during convert but not training

Hi, i train with the Dlight Trainer (without problems) but when i want to use the converter.... :? Can someone help her? Faceswap is up to date (2019.11.10)Win10 OS 11/10/2019 08:38:17 MainProcess MainThread config check_exists DEBUG Config file exists: 'C:\Users\denni\faceswap\config\convert.ini' 1...
by tochan
Fri Oct 11, 2019 4:33 pm
Forum: Hardware
Topic: Differnt GPU's in one System
Replies: 2
Views: 4463

Re: Differnt GPU's in one System

I do this but only the freqency from the 1080 go high an "a little bit the Temps". with select one GPU the 1080 freqency is in idle. Need i Linux for this?

by tochan
Fri Oct 11, 2019 4:15 pm
Forum: Hardware
Topic: Differnt GPU's in one System
Replies: 2
Views: 4463

Differnt GPU's in one System

Hi, last month my 1080 died so i buy a 2080 for the warrenty time ;). Now the fixed 1080 is back and i plug the 2 Cards in my Sys butt the Training speed is not faster or bigger batches are possible. The Clock speed of card 1 and 2 are high and not in the idl but the Temp from the second GPU is very...
by tochan
Sun Sep 22, 2019 11:23 am
Forum: Hardware
Topic: GPU Training Speed
Replies: 11
Views: 19718

Re: Hardware best practices

For Traing speed Information (Trainer Dfl-128):
1080 GTX EVGA Hybrid (AIO Whatercooler) 18,2 (12k Iterations)
2080 RTX MSI Aero (Blower Cooler)25,2 (15k iterations)