Review of image inpainting in practical challenge
Yuhang Shen · Applied and Computational Engineering · 2023
As a very popular computer vision direction in recent years, image inpainting has developed very rapidly and produced many valuable results, but there are still many problems that need to be solved. This paper focuses on classic model research, related data sets, and evaluation standards. Through the analysis of specific examples, it is found that the effect of the existing model on the inpainting of some unique pictures is still unsatisfactory, and a solution is proposed to optimize the specificity of the training set by modifying the training set so that the model and the corrupt image are more correlated, which will result in more relative results.