Questions of sparse view camera poses

[ghost]

Hello, I really appreciate your work. I don’t know much about the work in the field of sparse view 3D reconstruction. After reading your code and comparing the codes of other related works, I found that the camera poses of the training views and test views of sparse perspective 3D reconstruction are all used in the original data set, i.e. the camera pose calibrated using all views. Is my understanding correct? Thank you very much!

[zhongyingji]

I think you are correct. They use point cloud generated from sparse views, e.g., 12 views for mipnerf360, as initialization: https://github.com/VITA-Group/FSGS/blob/main/tools/colmap_360.py#L169.

But they use train/test poses by COLMAP from all views:
https://github.com/VITA-Group/FSGS/blob/main/scene/dataset_readers.py#L284-L297

BTW, I am quite curious about the coordinate systems that are used for the above two times. They are different, but the author just initialize the point cloud from sparse views, which does not lie in the same coordinate system with the camera poses.

Correct me if I am wrong🙏

[xuncpp]

I also think you are right. Is there any work currently being done to discuss and research this? I haven't seen enough work yet. Thank you！