Open kke19 opened 1 week ago
Hi @kke19, thanks for your interest!
Estimating point clouds (depths) from extremely sparse views can be difficult even with known camera poses. This usually requires generalizable network based methods. If you are targeting on objects, you can try the recent GeoLRM that infers points (gaussian splats) from multiple posed images. For general purposes, I think you can try UFORecon for sparse views, and the MVSNet family PointMVSNet, FastMVSNet, MVSNet for dense views.
Hope these methods can be helpful to your work!
HI, I am very grateful for your research and how it has helped my work. However, I have encountered a problem. I hope to generate sparse point clouds using the results produced by your research. In the past, I have used COLMAP to generate sparse point clouds, but now my research scenario has changed to sparse point cloud reconstruction under sparse viewpoints, and COLMAP has failed in this scenario. If you know how to reconstruct sparse point clouds using the sparse viewpoints poses produced by your research, please provide me with some guidance. Thank you very much.