You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
From the code, we can find that the matrix MtM is initialized as ones. MtM(rand(size(MtM)) < 0.5 ) = 1; means that the missing part in the mask is ones. Above all, the mask doesn't work, so the signal_sparse is same as signal. Then we can find that the inpainting process is same as the process of Convolutional Sparse Coding, it doesn't do anything with inpainting. Then, i initalize the MtM as zeros. The inpainting results of the code are bad, it is quite different from the results in the paper. I want to know how we can get the great inpainting results in the paper.
The text was updated successfully, but these errors were encountered:
reconstruct_2D_subsampling.m
From the code, we can find that the matrix MtM is initialized as ones. MtM(rand(size(MtM)) < 0.5 ) = 1; means that the missing part in the mask is ones. Above all, the mask doesn't work, so the signal_sparse is same as signal. Then we can find that the inpainting process is same as the process of Convolutional Sparse Coding, it doesn't do anything with inpainting. Then, i initalize the MtM as zeros. The inpainting results of the code are bad, it is quite different from the results in the paper. I want to know how we can get the great inpainting results in the paper.

The text was updated successfully, but these errors were encountered: