eduzhai > Applied Sciences > Engineering >

Progressively Unfreezing Perceptual GAN

  • Save

... pages left unread,continue reading

Document pages: 16 pages

Abstract: Generative adversarial networks (GANs) are widely used in image generationtasks, yet the generated images are usually lack of texture details. In thispaper, we propose a general framework, called Progressively UnfreezingPerceptual GAN (PUPGAN), which can generate images with fine texture details.Particularly, we propose an adaptive perceptual discriminator with apre-trained perceptual feature extractor, which can efficiently measure thediscrepancy between multi-level features of the generated and real images. Inaddition, we propose a progressively unfreezing scheme for the adaptiveperceptual discriminator, which ensures a smooth transfer process from a largescale classification task to a specified image generation task. The qualitativeand quantitative experiments with comparison to the classical baselines onthree image generation tasks, i.e. single image super-resolution, pairedimage-to-image translation and unpaired image-to-image translation demonstratethe superiority of PUPGAN over the compared approaches.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×