Page 1 of 1

Error: file folder contains no images

Posted: Mon May 29, 2023 12:33 pm
by papaquans

Hello:

I'm trying to train using a faceset that I got off the internet. I'm new to this, so this might just be something that I'm doing wrong.

The faceset that I personally scanned, works. Everything is smooth without problems. The faceset I downloaded off the internet, not working. It says Error: "filename" contains no images . Except, there are images. If I go into file explorer and search for the file folder, I can very well see the images there. I don't know what's going on here. Any quick suggestions?


Re: Error: file folder contains no images

Posted: Mon May 29, 2023 2:45 pm
by torzdf

You can only train with Faceswap extracted images. They contain required information in the PNG header.


Re: Error: file folder contains no images

Posted: Mon May 29, 2023 6:51 pm
by papaquans
torzdf wrote: Mon May 29, 2023 2:45 pm

You can only train with Faceswap extracted images. They contain required information in the PNG header.

Oh, is that so? Well, that's important information to me.

Although, if thats true, how can I use facesets that I got online? Is there a way to format those in a way I can use? Surely I don't have to rescan and personally format a new faceset each time I want to iterate a model right?


Re: Error: file folder contains no images

Posted: Tue May 30, 2023 10:11 am
by torzdf
papaquans wrote: Mon May 29, 2023 6:51 pm

Although, if thats true, how can I use facesets that I got online?

You don't.. Or, if you must, then you need to run extraction on the faceset so they come out at the correct size, with the correct alignment, containing the required information. Far better to generate your own sets though.

Is there a way to format those in a way I can use? Surely I don't have to rescan and personally format a new faceset each time I want to iterate a model right?

See above. The PNGs hold mask and header information that can only be generated by Faceswap's extraction process. The consistent alignment required is also generated by Faceswap's extraction process. Once extracted, these can be used to train as many models as you wish.