view article
Figure 7
A comparison of inpainting methods, where the masked input data are displayed in the top row, followed by their corresponding ground-truth images in the next row. The dashed areas in the ground-truth images indicate the location of the close-up images presented in Fig. 8. The following rows present the inpainted results obtained from the deep learning methods studied in this paper, where the non-gap regions have been replaced with the original pixel intensities from the input images. These results are organized according to the overall performance per method from the highest to the lowest, as follows: MSDNet, TUNet, partial convolution, convolutional autoencoder and biharmonic functions.

Journal logoJOURNAL OF
APPLIED
CRYSTALLOGRAPHY
ISSN: 1600-5767
Volume 55| Part 5| October 2022| Pages 1277-1288
https://doi.org/10.1107/S1600576722007105
Follow J. Appl. Cryst.
Sign up for e-alerts
E-alerts
Follow J. Appl. Cryst. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS