caiyuanhao1998 / SAX-NeRF

"Structure-Aware Sparse-View X-ray 3D Reconstruction" (CVPR 2024)
https://arxiv.org/abs/2311.10959
MIT License
469 stars 19 forks source link

Custom dataset #19

Closed leejngmn closed 2 months ago

leejngmn commented 2 months ago

Hello! I appreciate your research. Your research has led to some interesting collaborations.

The data we use is too big, so we are trying to reshape it to 256256256 size like the sample you provided. However, if we set it to 204820482160, an error occurs. What does the z-axis mean here? What should go in place of 2160?

In addition, are there any parameters to keep in mind when using a custom dataset?

leejngmn commented 2 months ago

This is a config based on Dicom data. Is there anything wrong? I wonder if the 2D-Pixel Size in X,Y direction is correct, and how can I find out the Accuracy of FWD proj value.

Thank you so much!

caiyuanhao1998 commented 2 months ago

Hi, thanks for your interest.

I suggest you first downscale the CT volume from 2048 x 2048 x 2160 to 256 x 256 x 270. The z-axis does not have any physical meaning. But I suggest you visualize the data to make sure it seems plausible, that is, you can tell it from the projection. Or you'd better rotate the axis until it looks plausible. Just follow our instruction about how to use TIGRE toolbox and use our data generation to synthesize the pickle file containing the 2D projections.

Hope this information may help :)

caiyuanhao1998 commented 2 months ago

The parameters of config should be set in the yaml file and the config setting of the data generation