Face completion with Hybrid Dilated Convolution

作者：

Highlights：

• We propose a face completion network combining the U-Net with HDC blocks.

• We propose linear-growth dilation to address gridding effect in face inpainting.

• We improve the training stability of our model with the spectral normalization.

• Our model can produce photo-realistic higher-quality images in face completion.

摘要

•We propose a face completion network combining the U-Net with HDC blocks.•We propose linear-growth dilation to address gridding effect in face inpainting.•We improve the training stability of our model with the spectral normalization.•Our model can produce photo-realistic higher-quality images in face completion.

论文关键词：Image completion,Image inpainting,Deep learning,GAN

论文评审过程：Received 9 February 2019, Revised 24 July 2019, Accepted 6 October 2019, Available online 14 October 2019, Version of Record 22 October 2019.

论文官网地址：https://doi.org/10.1016/j.image.2019.115664