Face Generation with NVIDIA StyleGAN2 ADA and Python 3 (7.3)

[@r00t_sh3ll]

Amazing, loving this Jeff. Thanks for the video and the code.

[@HeatonResearch]

You are most welcome, glad it is useful.

[@catherinele8417]

Ok i have the answer from my previous question asked in comments on the other video . Thanks for your videos !

[@NerdyRodent]

512x512 fits into 8gb vram, but you need to change the gpu batch size.

[@HeatonResearch]

Good point! Thanks

[@stephenhalker]

I'm really interested in the process of training a GAN, and found this video. did you take a more in depth dive into "how to train a GAN" in a previous video?

[@franciscog5261]

Hello! First of all, thank you for your work and for sharing it. I have a question: I am wondering if it is possible, by using a pre-trained dataset, get a styleGAN proyection (interpretation) of few real people portraits , and then, combine those interpretations into stylegan transitions? Since this would be based on a pre-trained model, no hi-end GPU would be needed. Have you made any tutorials about that already? Thank you!

[@HeatonResearch]

There is research on that, and I am reviewing that for a future video.

[@StreetCertifiedRecords]

How does an RTX3090 perform compared to a V100 all else equal?

[@DungPham-ai]

Thank you so much. Amazing ! Good job.

[@peterweyland117]

Hi Jeff. Thanks for the video. How to make video from images with morph transition?

[@PaulFishwick]

Jeff: the nvidia repo training phase creates Image collages (8x15) in the form of “fakes....N” and corresponding Pkl files with N in the file name. Do you know a way to get either the seed or vector if we see an interesting image (in the 8x15 collage) and want to locate the precise vector? This would allow some experiments post-curating the fakes.

[@cianakril]

Cut out the needed face from 8x15 and project it into relevant model snapshot.

[@lexscarlet]

Do you use any cloud resources like weights and biases or lambda compute or one of the many others I don't know or am misnomering?

[@HeatonResearch]

I do use cloud at times; however, this was all trained on hardware that Lenovo is letting me use for awhile.

[@lowqchannel]

can generate face using pretrained model with 8GB memory geforce gpu ?

[@gabrielsalvador1613]

Can someone tell me how to those sketch-to-real-face using this code?

[@sreejaysreedharan4085]

This is interesting..Thanks Jeff

[@syeedafatima8557]

How to build a pre trained model of ua own dataset. Please make a basic beginner tutorial

[@Micropterus06]

Jeff you're the best, thanks

[@HeatonResearch]

Thanks!

[@michaelantunes9397]

What are your thoughts on Pop OS?

[@HeatonResearch]

I have not worked with it, yet... I've seen it as an option on System76

[@ammarkhan2611]

Is it possible to train it on 1024X768 or 512X384 images

[@HeatonResearch]

Not without change to the NVIDIA source code for StyleGAN2, they require square.

[@ammarkhan2611]

Thank you for your reply, sir... Is there anything else (apart from stylegan2 or stylegan2ada )that can be used to generate 1024X768 images?

[@fernandosanchezvillanueva4762]

Great video !!!

[@HeatonResearch]

Glad you enjoyed it

[@markusepple6204]

7:05 left ear of the girl is too big.

[@HeatonResearch]

Agree

[@XX-vu5jo]

I wish he was my phd professor indtead of the clowns in my university

[@HeatonResearch]

Thanks!

[@XX-vu5jo]

@@HeatonResearch Is there a way for me to take a ph.d with you as my professor? I am planning to take another one from a real university.

[@HeatonResearch]

@@XX-vu5jo At this point, I do not take PhD students, I just teach one class as an adjunct. I am more on the industry side, I do hire interns, though. sites.wustl.edu/jeffheaton/internships/

[@XX-vu5jo]

@@HeatonResearch Do you accept lab assistants that can work remotely? Or we can just work on a research work or something together? Would love to learn from you. You are the real thing! I am so sick and tided with the fakers here. They even told me and the other phd students that CNN is not part of computer vision!