Image Inpainting Algorithm Based on Neural Network and Attention Mechanism
Li Yu, Xiuyuan Yu, Bao Qiliang · 2019
At present, there are obvious defects in the two mainstream generation models VAE and GAN of deep learning in the field of image inpainting.Based on the advantages of both, the paper recommends conditional information and attention mechanism, proposesing the ACVAE-PGGAN network model, which draws on the progressive growth training method and performs lightweight compression.The experimental result shows that the improved network model proposed in paper enhances training stability, which improves the resolution of generated images and the number of running frames effectively.