eduzhai > Applied Sciences > Engineering >

GL-GAN Adaptive Global and Local Bilevel Optimization model of Image Generation

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Although Generative Adversarial Networks have shown remarkable performance inimage generation, there are some challenges in image realism and convergencespeed. The results of some models display the imbalances of quality within agenerated image, in which some defective parts appear compared with otherregions. Different from general single global optimization methods, weintroduce an adaptive global and local bilevel optimization model(GL-GAN). Themodel achieves the generation of high-resolution images in a complementary andpromoting way, where global optimization is to optimize the whole images andlocal is only to optimize the low-quality areas. With a simple networkstructure, GL-GAN is allowed to effectively avoid the nature of imbalance bylocal bilevel optimization, which is accomplished by first locating low-qualityareas and then optimizing them. Moreover, by using feature map cues fromdiscriminator output, we propose the adaptive local and global optimizationmethod(Ada-OP) for specific implementation and find that it boosts theconvergence speed. Compared with the current GAN methods, our model has shownimpressive performance on CelebA, CelebA-HQ and LSUN datasets.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×