eduzhai > Applied Sciences > Engineering >

BiO-Net Learning Recurrent Bi-directional Connections for Encoder-Decoder Architecture

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: U-Net has become one of the state-of-the-art deep learning-based approachesfor modern computer vision tasks such as semantic segmentation, superresolution, image denoising, and inpainting. Previous extensions of U-Net havefocused mainly on the modification of its existing building blocks or thedevelopment of new functional modules for performance gains. As a result, thesevariants usually lead to an unneglectable increase in model complexity. Totackle this issue in such U-Net variants, in this paper, we present a novelBi-directional O-shape network (BiO-Net) that reuses the building blocks in arecurrent manner without introducing any extra parameters. Our proposedbi-directional skip connections can be directly adopted into anyencoder-decoder architecture to further enhance its capabilities in varioustask domains. We evaluated our method on various medical image analysis tasksand the results show that our BiO-Net significantly outperforms the vanillaU-Net as well as other state-of-the-art methods. Our code is available atthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×