Faceswap Forum

@Kvrooman did a fix for this but hadn't pushed to Master.

Just pushed. Update faceswap and try again

There was a bug in serial processing. I have pushed a bugfix, so please get the latest code and try again.

Let me know if the problem persists.

Yes, they are both heavy models. Just lower the batch size until it runs

If you haven't already, you should read the training guide, as it talks about selecting the batch size
viewtopic.php?f=6&t=146

The Best Practice tips also offer some pointers for eeking out more quality:
viewtopic.php?f=6&t=74

At the moment, not within Faceswap itself (but this is a feature we do want to add).

You can force which GPU by setting the Environmental Variable CUDA_VISIBLE_DEVICES

https://stackoverflow.com/questions/497 ... indows-cmd

Post your issue in the linux install thread, so others can see the issue.

The installer is now the correct way. Hopefully the video will get updated

The OpenCV and Pillow writers both have this option...

Edit > Configure Convert Plugins > Writer, then select "Draw Transparent".

You need to clean the alignments file. Whilst you have removed the faces, they still exist in the Alignments file. See here:
viewtopic.php?f=5&t=27#clean

My personal choice would be the 2080ti, purely because it reduces complexity, but ymmv

It looks like you have images that are not faces inside your faces folder.... Ultimately *all* images in the faces folder need to be the same size (as it returns the batch of images as an array). It looks like it has failed in instances where this is not the case.... Specifically (as an example, the...

That GPU is right at the bottom end of what will run.... In the first instance, enable the "Allow Growth" option. There is a bug in TF with some combinations of OS/GPU/Cuda (there appears to be no consistent pattern) where this is necessary. I imagine the next thing that will happen is you run out o...

I'm going to lock this as it won't be possible to diagnose further.

If you continue to have issues, please start another topic.

Thanks.

Can you try enabling the "allow_growth" option and see if that fixes the issue?

Honestly? Without a test video, I can't really diagnose. We plug into FFMPEG to mux audio, so it should work, but it may be a mapping issue.

InternalError: GPU sync failed

This is a hardware/driver issue. Most likely power or overclocking related.

https://www.google.com/search?q=Interna ... ync+failed

It looks like a problem with one of your images.

Are you able to zip up your facesets and share somewhere so I can see if I can diagnose the issue?

MemoryError

This means that you have run out of System Ram. Sorting faces is memory intensive (this is warned about in the guide).

Split your faceset up into smaller batches.

There is a conflict on your system between SciPy and some installed DLLs.

See here:

https://github.com/conda/conda/issues/6396

CUDNN errors sometimes occur on some GPU/OS/Driver combinations. There doesn't seem to be any pattern to it. They can normally be averted by enabling the "allow_growth" option.

We are planning to put better multi-gpu handling into Faceswap soon.

This is a Conda issue, not Faceswap. One of it's internal libraries is breaking.

I recommend following these steps to clean everything up and try again:
app.php/faqpage#f1r1