deep prior

[datseismo]

Hii, developers,

[weijia29]

Thanks developers for updating the codes.It runs even in cpu also for small volume of data.Nevertheless, the query asked by @datseismo is valid and I also tried with his data sets and I think reshaping/little modification in code is required to rectifying the problem.Developers can show the better path for the same.

[fpicetti]

Hi @datseismo and @weijia29,

Thanks for spotting the bug. Last commit should have fixed it.

The procedure you tested is correct, you need to have a (M, N, 1) dataset out of your (M, N) original dataset.
Regarding the 48-shape problem, I think it is due to the network dimensions (it try to shrink too much an already-small axes).
You can try either reducing the network filters and skip, or zero-padding your dataset, updating the mask subsequently.

I have uploaded a proof_of_concept_2D notebook, using the lines dataset. Check it out, and let me know if the problem persists.
Otherwise, you can have a look at the antialiasing_lines notebook, that is a slight modification of the method. There, you can find all the steps of the interpolation, without classes and hidden functions.