simoroma
Cannot get your results
Hey there, Thanks for your blog post and implementation. I have been trying to use your code to get z vector of my own images, however my results don't look as good as you no matter how many iterations I use. I have attached a sample celebA face and the reconstructed face I achieve. Can you help me with fixing my code? reconstructed_imgs
Does the image come from CelebA dataset? Could you try to use the same images I used in the blog post so you can double check if it works. Or you might also try to generate an image first based on this code https://github.com/tkarras/progressive_growing_of_gans Then you can try to recover the latent vector.
You could also try out a more recent version based on StyleGAN2 https://colab.research.google.com/drive/1ShgW6wohEFQtqs_znMna3dzrcVoABKIH
Did you use the same processing steps as in stylegan (e.g. blur background, center the face, etc) ?
Hello, I crop the faces using MTCNN.
Yours Sincerely, Nicky Bayat
On Feb 16, 2021, at 6:28 PM, Dat Duong [email protected] wrote:
Did you use the same processing steps as in stylegan (e.g. blur background, center the face, etc) ?
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.