search

caiyuanhao1998 / X-Gaussian

"Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesis" (ECCV 2024)
https://arxiv.org/abs/2403.04116
214 stars 5 forks source link

Questions about CT Reconstruction #4

Closed byungjur96 closed 3 weeks ago

byungjur96 commented 3 weeks ago

Hi, first of all, thank you for sharing your fascinating research. I'm experiencing amazing result with your method with CT reconstruction tasks. However, I'm currently stuck in saving the result 3D CT volume. Can you help me how to save the result 3D CT volume?

And one more, my result with Ellipse shows only white screen

image

Since the rendered image and initialization has been successfully done, I guessed that that's due to overlap of tremendous gaussians. Is it proper result compare to previous your experiments? Thanks a lot in advance for your reply!

caiyuanhao1998 commented 3 weeks ago

Hi, thanks for your interest.

Our X-Gaussian is mainly proposed for Novel View Synthesis. How to directly retrieve the CT volume is a difficult problem. For the CT reconstruction task. We firstly use X-Gaussian to generate projections. Then the synthesized and real projections are fed into traditional CT reconstruction methods to boost their performance.

If you want to directly reconstruct the CT volume, please follow our CVPR 2024 work - SAX-NeRF: https://github.com/caiyuanhao1998/SAX-NeRF/

If you find our repo useful, please help us star it. Thanks :)

byungjur96 commented 3 weeks ago

Then I'm wondering how did the Sparse-view CT reconstruction has been done Is it right that the SVCT reconstruction experiments has been done with 5 given images and 95 synthesized novel view images? Are these synthesized images are created only with 5 given images?

caiyuanhao1998 commented 3 weeks ago

Yeah. Directly retrieving the 3D CT volume from the Gaussian point clouds is a very complicated problem.

We have another paper to solve this problem. Please take a look at:

https://arxiv.org/abs/2405.20693

byungjur96 commented 3 weeks ago

Thanks for guidance! Since your wonderful research SAX-NeRF, I've been keep following your researchs and inspired a lot from them. I've been wondering when does the code of r2 gaussian will be released. By the way, thanks a lot again.

caiyuanhao1998 commented 3 weeks ago

Thanks for your recognition!

If you find our repo useful, please help us star it. Thanks :)