s-omidikia comments

Repositories
Issues
Comments

Results 3 comments of


                                            s-omidikia

All of the losses except the reconstruction loss don't change.

> For your medical dataset, is the generator trained by yourself? Thank you for your consideration First of all, we trained a conditional StyleGAN from scratch on our medical image...

All of the losses except the reconstruction loss don't change.

Yes, one-hot labels are fed to the mapping network, and the point is that the generated image during the GAN-inversion process is a realistic image, but it hasn't converged to...

All of the losses except the reconstruction loss don't change.

Thank you for your consideration. 1. The initial reconstruction wasn't realistic, and it took several iterations to make it more realistic. However, at a certain point in the training process,...