Frequency Domain Image Translation: More Photo-realistic, Better Identity-preserving
Image-to-image translation has been revolutionized with GAN-based methods. However, existing methods lack the ability to preserve the identity of the source domain. As a result, synthesized images can often over-adapt to the reference domain, losing important structural characteristics and suffering...
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Artículo |
| Publicado: |
2020
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2011.13611 http://arxiv.org/abs/2011.13611 |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|