eduzhai > Applied Sciences > Engineering >

SiENet Siamese Expansion Network for Image Extrapolation

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Different from image inpainting, image outpainting has relative less contextin the image center to capture and more content at the image border to predict.Therefore, classical encoder-decoder pipeline of existing methods may notpredict the outstretched unknown content perfectly. In this paper, a noveltwo-stage siamese adversarial model for image extrapolation, named SiameseExpansion Network (SiENet) is proposed. In two stages, a novel border sensitiveconvolution named adaptive filling convolution is designed for allowing encoderto predict the unknown content, alleviating the burden of decoder. Besides, tointroduce prior knowledge to network and reinforce the inferring ability ofencoder, siamese adversarial mechanism is designed to enable our network tomodel the distribution of covered long range feature for that of uncoveredimage feature. The results on four datasets has demonstrated that our methodoutperforms existing state-of-the-arts and could produce realistic results.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...