Search found 2704 matches

by torzdf
Mon Jun 17, 2024 5:42 pm
Forum: Hardware
Topic: Help on choosing a GPU for faceswap
Replies: 1
Views: 38

Re: Help on choosing a GPU for faceswap

Go for the newer card imho (although there won't be a lot in it).

Avoid the overclocked variants. Overclocking is bad for Machine Learning, so you would need to turn off the stock overclocks anyway.

by torzdf
Mon Jun 17, 2024 5:40 pm
Forum: Convert Discussion
Topic: Faceswap of someone who is wearing a mask
Replies: 1
Views: 26

Re: Faceswap of someone who is wearing a mask

Without really understanding the specifics, the short answer is 'yes', but it will likely depend on what you are actually trying to achieve. Related: https://github.com/deepfakes/faceswap/pull/1210

by torzdf
Fri Jun 07, 2024 4:58 pm
Forum: Extract Support
Topic: Extraction problem
Replies: 2
Views: 567

Re: Extraction problem

That sounds like an issue with the video file in some way.

What I would try is to convert the video file into frames (you can do this inside faceswap with the EFFMPEG tool), and then try to extract from the frames

by torzdf
Fri Jun 07, 2024 4:56 pm
Forum: Installation Support
Topic: Installation Problem conflicting Numpy Requirements
Replies: 1
Views: 725

Re: Installation Problem conflicting Numpy Requirements

Try the following:

open a Terminal emulator and then

Code: Select all

conda activate faceswap
conda install numpy">=1.24,<1.25"
by torzdf
Fri Jun 07, 2024 4:51 pm
Forum: Training Support
Topic: NSWindow should only be instantiated on the main thread
Replies: 4
Views: 771

Re: NSWindow should only be instantiated on the main thread

Hmmm. This looks very much Apple Silicon specific. Unfortunately I do not own a Mac to be able to test/troubleshoot.

I have recently put Faceswap 3 into beta, so if you are feeling adventurous, you could test that out and see if your issue persists: viewtopic.php?t=3010

by torzdf
Fri Jun 07, 2024 4:41 pm
Forum: General Discussion
Topic: Faceswap 3 open for testing
Replies: 0
Views: 1220

Faceswap 3 open for testing

Faceswap 3 is in Beta Over the past few months I have been busy updating Faceswap to use Keras 3/PyTorch as its backend. This is a major change, the reasons being that Tensorflow have made some baffling decisions of late that make it harder and harder to support going forward. This has been a mammo...
by torzdf
Wed Jun 05, 2024 12:45 pm
Forum: Extract Support
Topic: Intel Arc A550M GPU
Replies: 1
Views: 408

Re: Intel Arc A550M GPU

If it is supported, it will use the DIRECTML version of Faceswap, so make sure you are using that.

If you are then please provide the details as per: app.php/rules#rule-4b

by torzdf
Wed Jun 05, 2024 12:44 pm
Forum: Extract Discussion
Topic: Alignment question
Replies: 4
Views: 793

Re: Alignment question

You don't need to overthink/overdo it. If the extracted face looks aligned, call it correct and move on. The model has a certain amount of augmentation built in to deal with minor misaligns. For training sets I just delete any misaligned faces. For convert, I will fix obvious misaligns, and then run...
by torzdf
Tue Jun 04, 2024 1:23 pm
Forum: Extract Discussion
Topic: Alignment question
Replies: 4
Views: 793

Re: Alignment question

Yes. it would impact alignment. To what amount, I could not say.

by torzdf
Tue Jun 04, 2024 1:22 pm
Forum: Extract Support
Topic: Error generating mask due to invalid tensor format fp32
Replies: 3
Views: 3578

Re: Error generating mask due to invalid tensor format fp32

There are upcoming changes that will hopefully address this. Keep an eye out for an announcement in the coming week or so

by torzdf
Tue Jun 04, 2024 1:21 pm
Forum: Introductions
Topic: hello
Replies: 1
Views: 460

Re: hello

Welcome to the forums

by torzdf
Tue Jun 04, 2024 1:21 pm
Forum: Training Support
Topic: guidance on using multiple PC's simultaneously towards the same project (poor mans render farm)
Replies: 1
Views: 497

Re: guidance on using multiple PC's simultaneously towards the same project (poor mans render farm)

This is not currently implemented in Faceswap, so you would need to go into the code.

It is something I may look at in the future though

by torzdf
Sat May 25, 2024 2:30 pm
Forum: Extract Discussion
Topic: Using filter/nfilter for alignment file
Replies: 5
Views: 1176

Re: Using filter/nfilter for alignment file

Profiles are notoriously tricky....

If you would rather deal with more false positives, but ultimately have more correct profiles, then you can turn off most/all aligner filters (I'd probably leave the 'features' filter enabled, as any which get picked up by that will definitely be misaligned)

by torzdf
Fri May 24, 2024 11:59 am
Forum: Extract Discussion
Topic: Using filter/nfilter for alignment file
Replies: 5
Views: 1176

Re: Using filter/nfilter for alignment file

It will, yes.

I tend not to use the filter/nfilter. They are fine, it just tends to be quicker, in the long run, to extract all faces, then use "sort by face" and remove all of the unwanted faces that way.

Less risk of false positives/negatives too.

by torzdf
Fri May 24, 2024 11:44 am
Forum: Extract Discussion
Topic: Using filter/nfilter for alignment file
Replies: 5
Views: 1176

Re: Using filter/nfilter for alignment file

The identity test is done after alignment, so it will not save you extraction time, no.

It has to work this way, because it can only test if the face is a particular identity, once it has an aligned face to test.

by torzdf
Fri May 24, 2024 11:42 am
Forum: Hardware
Topic: Supernoob, please guide !
Replies: 3
Views: 1102

Re: Supernoob, please guide !

Model output size, rather than final frame size. You may be able to get up to 256px face patches with that setup, but when learning the process you are going to want to work with lower resolution output anyway (64/128px) as models at this size train quicker, so you will learn what works and doesn't ...
by torzdf
Fri May 24, 2024 11:40 am
Forum: Convert Support
Topic: convert error. train extract crash
Replies: 4
Views: 1244

Re: convert error. train extract crash

ben wrote: Thu May 23, 2024 3:01 pm

extract error
2024-05-23 17:59:53.937751: F tensorflow/c/logging.cc:43] HRESULT failed with 0x887a0005: done_event.fence->Signal(done_event.fence_value)

See here:
viewtopic.php?p=8706#p8706

by torzdf
Thu May 23, 2024 2:12 pm
Forum: Convert Support
Topic: convert error. train extract crash
Replies: 4
Views: 1244

Re: convert error. train extract crash

Your model file is corrupted. More info:
search.php?keywords=bad+object+header+version+number

by torzdf
Thu May 23, 2024 11:48 am
Forum: Hardware
Topic: Supernoob, please guide !
Replies: 3
Views: 1102

Re: Supernoob, please guide !

It is not an ideal setup, no. However it should work just fine, but will likely be a bit slower than a more modern setup, and you will be limited to lower resolution models.

However, if you are just looking to learn the process, then this should work ok for you.

by torzdf
Thu May 23, 2024 11:46 am
Forum: Extract Discussion
Topic: Optimal source for high quality file extraction
Replies: 1
Views: 711

Re: Optimal source for high quality file extraction

Avoid jpeg compression where you can. If you must compress use lossless (like PNG). From video sources, bitrate is hugely important. Sites like Youtube crush the bitrate hugely so are not suitable sources for the bulk of a training set. Avoid HDR content at all costs . HDR content does not train well.