Looks like a bug in the model.
Can you provide the full crash report please (linked at the last line of the output error)
Ok, looks like there are a couple of things going wrong here.
In the first instance, could you please follow all of these steps and try again:
If you hit an error again, please reply with your latest crash report.
I'm not sure what you mean. What's your end goal? What are you trying to achieve?
It's not possible to skip alignment as alignments is required to be able to extract the faces in the first place.
This sounds totally insane, and I'm surprised you got anything out of it!
But it's also interesting stuff, and I'm always keen to see results of people's esoteric experiments when they go outside of the norm. If there are any images you can share, please do, but appreciate that may not be possible.
More than that, it really depends on what you want to be looking at. Nearest neighbour will give you a (close to) pixel accurate output of what the model is doing. Bicubic will give you an output closer to what you are likely to see in the final convert.
It is teaching a model to learn how to upscale. Several models already work like this, the most obvious one being 'dfaker' off the top of my head.
Honestly, no idea. Changed loss configuration, changed augmentation settings, different BS is just impacting loss in a different way.
Ultimately, don't worry about it, as long as it is trending down.
Ok, no obvious issue there....
Do the same as above to get up an Anaconda prompt and do
Code: Select all
conda activate faceswap cd faceswap python faceswap.py gui
Hopefully we can get the actual text of the error message.
If no alignment is found then, in all likelihood, you are pointing convert at the wrong alignments file.
Make sure you have the file that contains every face in the video you are trying to convert from. See here:
If you can get away with not using mixed precision, do, as I had NaN issues with SYM-384 using MP. It resolves nice and quickly but NaNs earlier than I would have liked. Otherwise, I would say start with a fairly low lr. Maybe 3.5e-5, may 2e-5. Can't give more guidance beyond that sadly.
It would be very challenging. Personally I would probably avoid it. It would be easier the other way around.