Mr DeepFakes Forums
  • New and improved dark forum theme!
  • Guests can now comment on videos on the tube.
   
  •  Previous
  • 1
  • ...
  • 44
  • 45
  • 46(current)
  • 47
  • 48
  • 49
  • Next 
DeepFaceLab Explained and Usage Tutorial
(02-11-2019, 08:39 AM)iperov Wrote: You are not allowed to view links. Register or Login to view.
(02-11-2019, 08:31 AM)frosty3907 Wrote: You are not allowed to view links. Register or Login to view.It's 25GB... Is there something else I can check myself maybe?
This is from a model that's been training for a long time and was working previously up until the last two jobs.
upload only model files and data_dst root images which have a glitch in conversion,
and data_dst\aligned images convert to jpg via DFL .bat , also zip them

@frosty3907 please contribute to DFL, I have to check this bug.

OK I'll PM you a link with model, and original files of the selection of images previously shown.
(02-11-2019, 09:53 PM)dpfks Wrote: You are not allowed to view links. Register or Login to view.
(02-10-2019, 05:51 AM)Sentry79 Wrote: You are not allowed to view links. Register or Login to view.I'm on a gtx 970. I get crashes on my training unless I lower the batch size to 8. is this normal?

I also installed the latest drivers. I read earlier that I'm not supposed to install CUDA, since DFL already comes with it. Is that right?

What model are you trying to train? try reducing the batch size, and using H64 model to train.
I'm using H64 model. Any higher than 8 seems to make it unstable for me. I was wondering if this is normal for a gtx 970.
(11-15-2018, 10:18 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 12:53 PM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 11:51 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 02:37 AM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.
(11-13-2018, 09:09 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.I did yes. And I have a fine merged collection. But the "convertet to mp4" just shuts down instantly. 
I've used VirtualDub to make an avi, but it would be nice if DFL could convert it to mp4 with sound

I think it would be best if you open the bat file with a text editor (right click> open with>notepad or right click>edit) and add pause at the end, on a newline so the window will stay open and you will be able to see any errors if there are any.
I helped to spot the problem.
I gives the message:
[image2 @ 0000026576bf8600] Could find no file with path 'C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png' and index in the range 0-4
C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png: No such file or directory
Press any key to continue . . .

Did you delete any merged frames? (in particular frames 1-9)

It was something like that. My pics started at 00028. So I got a program called Bulk Rename Utility and renumbered all files from 00001 and it then worked again. Thank you!

Im working on this bulk renaming, i cant seem to get it done in a meaningful way. any tips.
is it possible to outsource a pornstar aligned faceset but not from the movie i chose that i want the celebrity to be based on ?

for example if i want to fake Stephanie Mcmahon on Kendra Lust's movie i have but i wanted to faster things up and got a premade of facesets of Kendra Lust from someone but the facesets are not from the movie im gonna fake Stephanie on.

i wonder if its possible?
(02-12-2019, 04:21 AM)blindboy Wrote: You are not allowed to view links. Register or Login to view.
(11-15-2018, 10:18 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 12:53 PM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 11:51 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 02:37 AM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.I think it would be best if you open the bat file with a text editor (right click> open with>notepad or right click>edit) and add pause at the end, on a newline so the window will stay open and you will be able to see any errors if there are any.
I helped to spot the problem.
I gives the message:
[image2 @ 0000026576bf8600] Could find no file with path 'C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png' and index in the range 0-4
C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png: No such file or directory
Press any key to continue . . .

Did you delete any merged frames? (in particular frames 1-9)

It was something like that. My pics started at 00028. So I got a program called Bulk Rename Utility and renumbered all files from 00001 and it then worked again. Thank you!

Im working on this bulk renaming, i cant seem to get it done in a meaningful way. any tips.

Not really on topic. But you can just use command line arguments to bulk rename. What exactly do you want achieved?
I'm working on a couple projects to get the ropes, last run doing SAE with 10.0 / 10.0 until 20K epochs and 0.1/4.0 until 50K and 0.1/0.3 until unlimited. Getting fairly good results.

This is guessing on my part. What are normally the triggers (apart from artifacts) to set the style power down? For now I'm just letting it run and will do smaller incremental runs in optimal projects later.

What other updates to the settings should I try at what moments? (using unobscured faces and similar face angle of the src and dst). Feed me your experience and I'll create a guide with the input and my own experience added.

Mostly looking at the image selection / training process for now. Optimising converting and post processing are for later.
(02-12-2019, 04:21 AM)blindboy Wrote: You are not allowed to view links. Register or Login to view.
(11-15-2018, 10:18 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 12:53 PM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 11:51 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 02:37 AM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.I think it would be best if you open the bat file with a text editor (right click> open with>notepad or right click>edit) and add pause at the end, on a newline so the window will stay open and you will be able to see any errors if there are any.
I helped to spot the problem.
I gives the message:
[image2 @ 0000026576bf8600] Could find no file with path 'C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png' and index in the range 0-4
C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png: No such file or directory
Press any key to continue . . .

Did you delete any merged frames? (in particular frames 1-9)

It was something like that. My pics started at 00028. So I got a program called Bulk Rename Utility and renumbered all files from 00001 and it then worked again. Thank you!

Im working on this bulk renaming, i cant seem to get it done in a meaningful way.
Can you just re extract the DST faces?
this problem happened to me today

all sudden DFL stopped working

when i try to extract data_dst it says extractor found 0 images and faces
(02-11-2019, 10:00 PM)yukinoyoru Wrote: You are not allowed to view links. Register or Login to view.
(02-11-2019, 09:53 PM)dpfks Wrote: You are not allowed to view links. Register or Login to view.
(02-10-2019, 05:51 AM)Sentry79 Wrote: You are not allowed to view links. Register or Login to view.I'm on a gtx 970. I get crashes on my training unless I lower the batch size to 8. is this normal?

I also installed the latest drivers. I read earlier that I'm not supposed to install CUDA, since DFL already comes with it. Is that right?

What model are you trying to train? try reducing the batch size, and using H64 model to train.

(02-11-2019, 09:07 PM)yukinoyoru Wrote: You are not allowed to view links. Register or Login to view.Hi i just trained my first model and created my first video, now i want to use the same source but a different destination, do i have to delete the model and train again from scrath? (0 epooch) or can i contrinue to train? (with my current epooch)

You can reuse the mode. Back it up, then just change the data_dst contents and re-train.

thanks, can i do that if i change the source as well or just if i change the destination?

thanks, can i do that if i change the source as well or just if i change the destination?

Yes you can re-use if you change the source as well. BUT in general, I would recommend you starting a new model unless the facial structures of the model you're trying to re-use is similar to the new data_src faceset.

(02-12-2019, 04:21 AM)blindboy Wrote: You are not allowed to view links. Register or Login to view.
(11-15-2018, 10:18 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 12:53 PM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 11:51 AM)emsbabs Wrote: You are not allowed to view links. Register or Login to view.
(11-14-2018, 02:37 AM)RandomAnon Wrote: You are not allowed to view links. Register or Login to view.I think it would be best if you open the bat file with a text editor (right click> open with>notepad or right click>edit) and add pause at the end, on a newline so the window will stay open and you will be able to see any errors if there are any.
I helped to spot the problem.
I gives the message:
[image2 @ 0000026576bf8600] Could find no file with path 'C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png' and index in the range 0-4
C:\DeepFaceLab\DeepFaceLab\workspace\data_dst\merged\%5d.png: No such file or directory
Press any key to continue . . .

Did you delete any merged frames? (in particular frames 1-9)

It was something like that. My pics started at 00028. So I got a program called Bulk Rename Utility and renumbered all files from 00001 and it then worked again. Thank you!

Im working on this bulk renaming, i cant seem to get it done in a meaningful way. any tips.

Why are your merged images missing in the first place? It's easier to just re-convert so it regenerates merged images?

(02-12-2019, 09:06 AM)sabrekism Wrote: You are not allowed to view links. Register or Login to view.is it possible to outsource a pornstar aligned faceset but not from the movie i chose that i want the celebrity to be based on ?

for example if i want to fake Stephanie Mcmahon on Kendra Lust's movie i have but i wanted to faster things up and got a premade of facesets of Kendra Lust from someone but the facesets are not from the movie im gonna fake Stephanie on.

i wonder if its possible?

Yes this is possible, but the faceset better be really good to include every angle to make sure the model is trained. Most people DO NOT do this because if we build a faceset from the data_dst video, it is guaranteed to have the angles we need in the faceset. Also, why train extra images that won't be used?

(02-12-2019, 06:51 PM)sabrekism Wrote: You are not allowed to view links. Register or Login to view.this problem happened to me today

all sudden DFL stopped working

when i try to extract data_dst it says extractor found 0 images and faces

Make sure you splice he data_dst.mp4 video into it's frames. This will generate images in data"dst folder. You need to make sure your PNGs are in this folder. Once you run the extractor, it will take these images and extract faces into data_dst\aligned
@dpfks thx for time boss

for some reason i still have the problem i re-installed the program many times and same thing happen

i get a message "extractor" didnt find any images "0" and "0" face detections

im worried
  •  Previous
  • 1
  • ...
  • 44
  • 45
  • 46(current)
  • 47
  • 48
  • 49
  • Next 

Forum Jump:

Users browsing this thread: monmon, Regit, 3 Guest(s)