Volume Reconstruction
byungjur96 opened this issue · 2 comments
Hi! First of all, thanks for sharing code of this wonderful work.
I have some questions about saving SR result as a volume file.
With default setting, it seems like it only saves only N_eval number of slides.
Is there any guidelines for saving real volume data? for example, npy file or nii.gz file again.
In my case, I upsampled 128^3 volume in 512^3 for SR with scale 4.0, so I put 512^3 bicubic upsampled volume for input.
However, ours.nii.gz file shaped like 512x512x50.
I hope I can get your help soon. Thanks in advance!
Hello. If i understand your question correctly, you can leave N_eval blank. And with the current repo, it will just store all 512* 512*512 images.
Thanks a lot! With more study of your code, I manually fixed N-eval=512 and got similar result.
Further, I have few more questions, and if possible, it would be really greatful if you help me dealing with some problems.
-
I'm running experiments with my personal data, and my experiments results in some voxel values with larger than 1, while all the inputs were normalized with the range between 0~1. Is there any normalization process during training? What is the expected range of result from your experiments?
-
In case of bicubic interpolation experiment you've used as an comparison method, I'm wondering what specific method are you using. I'm currently dealing with 3D SR tasks, and I'm still wondering what are 'cubuic' interpolation mean in 3D datsets, and I still didn't found solution.
Hoping for your replies.
Thanks a lot in advance!