Real world test issue

[syt06007]

Hello,

I really appreciate your open source code.

I have encountered a few issues while using the code. I have some questions regarding the real-world test. Are there specific hyperparameters or settings that you used? I am having trouble with the reconstruction using the Stanford dataset (Lego Knights and Bulldozer) as described in your paper.

I downloaded the data from this link: http://lightfield.stanford.edu/lfs.html. After obtaining the "Rectified and cropped images," I used a 17x17 image and input only the central 9x9 view into the model(crop, patchsize=128).

Thank you.