Question about training process

[dragonylee]

Thank you for your work. After reading the paper, I have a question regarding the training process that I would like to ask.

You had mentioned in your paper “For supervision, ground-truth intensity values can be obtained from the ground-truth CT image based on the coordinates of points by trilinear interpolation.” This implies that 3D CT data is needed as supervision during network training. Doesn’t this contradict the concept of "sparse view 3D reconstruction"?

I’m not sure if I misunderstood something, but I hope to get clarification. Thank you!

[lyqun]

3D supervision is necessary during the training as the method is data-driven. However, the 3D is not needed during the inference, and the network can reconstruct the 3D from sparse-view inputs.