Morphing NVIDIA StyleGAN2 ADA Pytorch Between Real Photos your GAN Doppelganger
HTML-код
- Опубликовано: 7 янв 2025
- Did you know you can find your GAN Doppelganger in the latent vector space of NVIDIA StyleGAN2 ADA? I provide a Colab notebook that contains all that you need to create your own morph videos between any two photos of people's faces.
2:12 Opening the CoLab Notebook
2:38 Use a Free Google GPU
2:52 Running the CoLab Notebook
3:27 Uploading the Starting/Stop Image
4:04 Installing StyleGAN2 ADA
4:15 Preprocessing Images for StyleGAN2
5:21 Converting Images to StyleGAN2 Latent Vectors
7:19 Building the Morph Video
8:17 Conclusion
Notebook used in this video:
github.com/jef...
Generating GANs:
github.com/jef...
Deal with CoLab timeout:
stackoverflow....
Download Images from Flickr
github.com/jef...
Follow Me/Subscribe:
/ heatonresearch
github.com/jef...
/ jeffheaton
Support Me on Patreon: / jeffheaton
![[1hr Talk] Intro to Large Language Models](/img/1.gif)
Hello.
Wow! I was waiting for that video for months! Quite likely you are the first person to implement this in colab, or at least in making it public.
This can be also done by installing stylegan locally, but it is so much easier to simply use colab.
Thanks a lot for sharing your work, very interesting.
I will be looking forward for this code explained.
Thanks again :)
Thanks, yes I usually get it working on my local Linux system, then try to demo it on CoLab for RUclips, if I can fit it into CoLab.
Excited for your future graphic machine learning ! really learn a lot from your videos.. thank you
Happy to hear that! More will be coming. Of course, make sure to subscribe so you do not miss anything! :)
great video! would be great to see the deep dive and how to make work that transition with a sequence of images
Yes , more detail on the code. Grest video Thanks
Great tutorial and great Google colab code ! The morphing is really smooth and realistic. I subscribe to see other good stuffs like that
Thanks! Welcome to the channel.
Hello
I am looking for hint for an algorithm that can morph simple shapes like a triangle into a square or two circles into one circle.
I would be thankful for any directions or links.
Great video, exactly what I have been looking for! Straight and to the point, and the result is great. Thank you for sharing all your work with the internet. So appreciated
thanks for this jeff ! does anyone have any slightly more updated versions of how you might do this?
you are great sir
You are giving extraordinary knowledge to students
Thanks! I have alot of fun in the process.
How can I do this for other set of images that are not faces?
is there a way to batch do this with a series of images?
Awesome, Thanks for sharing Jeff!
Superb. we need more tutorials like this.
Thank you, I will. Make sure to subscribe so you do not miss any!
Thank you so much! Amazing channel
deep dive into code &execution pls
For sure.
너무너무!! 고마워요!!!
THX~ so~~~~!!! much~~!!
Hi!
I have been testing your morphing notebook for a while, and I have a couple of questions, hope you don't mind.
It seems like it is difficult to get a projection that looks similar pictures of small kids. Could it be because the training set used, was trained mainly with faces of adults?
Also I tried to use the face of Homer Simpson, but the face detector won't recognize a face there. Is it possible to modify the face detector, to be able to morph between those cartoon faces?
And last, I guess it shoud be possible to morph between two different shapes, such as, the projection of a face, and the projection of, let's say, a butterfly.
Having the two projections: would the process be the same as done when morphing between two different face projections on this notebook?
Thank you, and thanks for sharing your research work. Looking forward for more morphing tutorials :)
I always get error when I go to upload an image
Fantastic!
Thanks Jeff. I’m currently trying to do something similar with a vae.
Good luck! Sounds good.
Really nice work and very interesting content. I'm wondering if it would possible to use GANs to combine (let's say) 4 faces and generate a new one out of them?
After setting up stylegan2 (thanks for the video by the way) I wrote something similar to this. I made the code go through several latend vectors. I noticed that times half way between some faces holes could appear in noses or weird growth would cover faces whilst in between faces. I think I know why but I don't know how to explain it in English. 😆
Does this compute on client side or server side @ Nvidia?
Amazing video! If you are using your own pre-trained data which is not human faces will you need to something different for the pre-processing? If so have you done any videos on how to go about that? Thanks Jeff, your videos are so helpful!
To what direction are you headed with these? I mean, what's a more meaningful uses for this aside from entertainment?
Its been a common request. It also shows how to map a real image to the GAN latent vector space.
Awesome, when are you uploading you going through the code
This is amazing! Do you think you could make a notebook that can modify some of the latent vector's (i.e. the stylegan doppelanger's) features? I have seen some repos and papers explore the latent space dimensions allowing them to modify things like gender, hair, ethnicity, etc.
So Cool Edition 😊 😊 😊
Thanks!
way more useful than my Asda toilet paper that broke on me.
please make a video on code explaination of this video
Dear Jeff.. does that mean you cant morph one of your gan-fishes into another? only human faces?
Great stuff btw.. Currently preparing my dataset to use it with your code ⚽
Yes I can morph one fish to another. But if I used the fish GAN as the starting point, and used Thor and I, we would both be projected into fish GAN space, which would not be optimal.
Nice work but it's time consuming and after waiting for an hour it returns from starting
Yes, you are right
Hail THOR ^^ We know who you are ^^
Hah! By the way, is anyone else looking forward to season 2 of Ragnarok? Binge watching season 1 is at least partly to blame for my choice of images.
@@HeatonResearch My mum ha
I love Thor, My Favorite Avenger 😇 😇 😇
Hah, yes me too. I did try to do Elon Musk to Tony Stark, but the GANs could not deal with Tony's facial hair pattern.
lol 4:25