Faceswap Forum

Ok, this issue is coming from Tensorflow. You can Google around to see if you can find a solution, but I suspect it *may* be a memory error. However, it's odd, because it appears that S3FD runs, then FAN fails (and S3FD is a LOT heavier than FAN). First ensure that any GPU overclocks are disabled Tr...

You only have 11 faces in your "B" side.

This isn't enough to train a model (see here: app.php/faqpage#f3r0)

Either way, there are a total of 14 samples per side to show the preview. As only 11 faces are being fed in, it errors.

That's fair enough. Like I say, I'll take another look. I can't see how this would have changed though, as the functionality hasn't been amended. Incidentally, I believe (and it's a long time since I wrote it) that the logic behind it was to avoid people accidentally pointing at the wrong faces fold...

Post the latest crash report please.

It looks like you are not training on extracted faces:

The contents of D:\Downloads\fs\fs2 should be faces extracted at the extract stage:
viewtopic.php?f=5&t=27

You can open the file and paste the contents into a message, or onto a site like pastebin and link it here.

Nothing has actively changed in this part of the code for quite some time. In fact I don't think I've made any changes to the alignment tool code for at least a year.... I will admit that your use case is not a use case that I ever envisaged, or used it for, but I will look into this when I get a ch...

When you manually fix the mask, any generated masks are erased (they have to be, as the masks no longer correspond to the actual face).

It's not ideal, but you can regenerate masks with the "mask" tool. If you select the mask(s) you want, select "missing" then it will fill in the gaps.

You can ignore the first part of the error, that's just a minor GUI display bug.

Can you hit the "Generate" button for Training and post the ouput command here?

Thanks

Ok, if you have "allow_growth" set to True in extract settings, this should now pull through for sort by face.

You *may* be able to train the Lightweight model on that card. It is specifically written for 2GB GPU cards, but it will be dependent on what else is using the GPU. Try it with a very low batchsize (2) and if it trains, you can try upping the batchsize in increments. If it doesn't train, try memory ...

The batch is split between GPUs.

To get the benefit of multiple GPUs, you would want to up the batchsize (i.e. if you are training BS 64 on 1 GPU, you'd want to train BS 128 on 2 GPUs).

Most likely memory issues. Probably on the edge of what your GPU can handle, hence the error.

If other models train without this error, then I would say that is definitely the cause.

You can try Memory Saving Gradients and/or Optimizer Savings to try to fit it in VRAM.

I will try to push an update for sort which reads the "ALLOW_GROWTH" from the extract config file.

If you don't see an update in the next 24 hours, please reply to remind me!

The first error: Running out of system RAM. The second error: Allow Growth issue + ran out of system RAM The 3rd error: Allow Growth issue. Unfortunately we do not have an "allow growth" option for sort, but you can still enable it by setting an environment variable. See here: https://forum.faceswap...

filter/nfilter is slow and not recommended. It's better to use "sort by face" afterwards.

This may be a bug, or it could have just hung for some reason.

This is just a GUI display bug where it tries to delete an object which doesn't exist.

It doesn't effect the running of the program, so hasn't been prioritized for finding.

Not really. The error reports are just a debug log output at the point of failure + some system information.

Generally it''s a question of looking at the Traceback and then working backwards to see what would have caused it to fail.

I haven't seen this before, It won't be anything specifically Faceswap, but some of the upstream libraries we use may be creating temporary files on the c drive?

Most likely you are out of GPU Memory (https://github.com/plaidml/plaidml/issues/225)

Try lowering the batch size.