Why do you use FFT and other images transformation?

[tempdeltavalue]

I saw same thing here,

https://github.com/greentfrapp/lucent
and
https://github.com/Mayukhdeb/torch-dreams/commits?author=Mayukhdeb

And I'm not really sure what I understand that...

Sorry to bother you, just curious 🙌

[eps696]

i'm afraid, i didn't quite understand your question.
confirmed [that's written in the very beginning of readme] - this generation method is based on the fft/ifft transforms, taken from the Lucent library (torch dreams by Mayukhdeb appeared later). why? just because ::]

on a seirous note - other CLIP-based text-to-image methods usually employed fat pretrained GAN generators, while i believed comparable [or better] results could be achieved with such lightweight ops as fft/dwt - if only i understood it enough to use and configure properly..

[tempdeltavalue]

My question was why you need fft for image generation at all ... but you already answered :)

[tempdeltavalue]

Probably interesting video to check in current context

https://www.youtube.com/watch?v=Al6NTbgka1o

[eps696]

yeah, good video, thanks.
the method itself has been published by another Matthew 2 years ago
[and also used in the very first tool from this repo - SIREN + FF mapping]