GAN2Shape icon indicating copy to clipboard operation
GAN2Shape copied to clipboard
verified publisher icon XingangPan

Try on AFHQ dataset

Open yzliu567 opened this issue 4 years ago • 2 comments

Hi, your ideas on this work are very nice. I'm trying on afhq dog dataset. I've trained stylegan-ada on this dataset and get the inversion. Then I feed the data into your pipeline and use the default mvn light file to train your model. But the result is not good. Could you please offer me some suggestions? (I took out 10 images from the dataset to train.) image image image image

yzliu567 avatar Jul 11 '21 11:07 yzliu567

I suggesting first performing joint pretraining on more samples (like 200 samples), and then train for each instance. This is what we did for face data.

XingangPan avatar Jul 25 '21 03:07 XingangPan

@yzliu567 Hi, I did a similar study, and I want to ask you a question please. I've trained stylegan2-ada https://github.com/NVlabs/stylegan2-ada on my own datasets and get the model (.pkl). But I don't know how to get the inversion (.pt). I think you have solved it yet. Looking forward to your reply. Thanks!

hito-Chen avatar Sep 02 '21 03:09 hito-Chen