The official code has been released now, you can find it here: github.com/XingangPan/DragGAN/tree/main
@anaanab Жыл бұрын
will u post a video on how to install and run it?
@kennette2121 Жыл бұрын
yeah man, you are one of the best out there so far that I've found! Super interested in the AI space although I am super busy with my primary job as a more full stack developer. Your videos keep the content clear and help me keep up with all the rapid movements in the space! Keep it up !
@letianwang5141 Жыл бұрын
awsome video! Both technical and understandable!
@Omnificuser Жыл бұрын
Keep going bro you have been a huge inspo for me!
@74Gee Жыл бұрын
So glad you're back!!
@martin-thissen Жыл бұрын
Thanks, happy to here I was missed! :-)
@t.b.5493 Жыл бұрын
Awesome video! Continue like that!
@martin-thissen Жыл бұрын
Thanks a lot! :-)
@Fredi-yu3rc Жыл бұрын
Love your videos! Could you make a video about the RedPajama-INCITE models?
@martin-thissen Жыл бұрын
Thanks so much! I will add it to my list. :)
@sidarth404 Жыл бұрын
Can you please solve the colab tortoise problem " To access the website, please confirm the tunnel creator's public IP below." that you provided.
@74Gee Жыл бұрын
I'm wondering if an interface that overlays a point mesh would significantly slow down the image generation. So using depth estimation to create the mesh and then a 3D package to manipulate the mesh to create the pixel vectors. When that is then passed to the image generation, would say 30 points have a 30 times increase in actually generating the image? Or would it have a similar generation time? My thinking is the point mesh could be more easily constrained to the form of the subject (face, body etc) as for these example the constraints are well known. The benefit of this would be the ability to drag say a nose out of bounds and it would manipulate the entire mesh to remain within bounds, maybe to turn the entire head. I suspect that adding multiple drag vectors would have some effect on generation time but possibly not. What do you think?
@martin-thissen Жыл бұрын
Yes, the more points you add the longer does the optimization of the intermediate latent w take. Actually the authors did exactly what you described in their paper and in their shown example it worked really well. Have a look at figure 7 in the paper if you want :) vcai.mpi-inf.mpg.de/projects/DragGAN/data/paper.pdf
@74Gee Жыл бұрын
@@martin-thissen Thanks that's an interesting paper. I should have read that first!
@velo1337 Жыл бұрын
when will you release a llama 2 video?
@martin-thissen Жыл бұрын
Appreciate that you're asking, just uploaded the video: kzbin.info/www/bejne/jaumhGuwbqykftk Hope you like it :-)
@Fredi-yu3rc Жыл бұрын
Is there a discord server for this channel? if not it would be great to have one!
@martin-thissen Жыл бұрын
I haven't created one yet, but am planning to do create one soon :)
@Fredi-yu3rc Жыл бұрын
@@martin-thissen That would be so great so people can help each other etc
@deeprajmajumdar14234 ай бұрын
how to place custom images of our own? and use that in this model?
@martin-thissen4 ай бұрын
Unfortunately, this is a little more complicated, but it can be done using GAN inversion: arxiv.org/pdf/2101.05278
@deeprajmajumdar14234 ай бұрын
@@martin-thissen would you create a video for that thanks in advance ☺️
@DeeprajMajumdarM23CSE0124 ай бұрын
@@martin-thissen Yes sir! Waiting for a video.
@bgNinjashows Жыл бұрын
KZbin algorithm is brutal my dude took some time off and algorithm definitely did not like that. Geez
@martin-thissen Жыл бұрын
Looks like that haha. But I will recover from it! :)
@bgNinjashows Жыл бұрын
@@martin-thissen 😀
@williamstrickland6892 Жыл бұрын
😴 Promo>SM
@leemark7739 Жыл бұрын
Not official implement.by the way can we process our own image
@martin-thissen Жыл бұрын
Yes, mentioned that in the video as well. The official code got released now: github.com/XingangPan/DragGAN/tree/main Yes, for this you need GAN inversion.