search

mkazhdan / PoissonRecon

Poisson Surface Reconstruction
MIT License
1.59k stars 431 forks source link

The reconstruction is not very good #273

Closed glt-mhs closed 1 year ago

glt-mhs commented 1 year ago

PoissonRecon --in p1.ply --out p1out.ply --depth 10 This is my reconstructed command, the same problem appears on every point cloud data, the effect is good in one direction, the effect is poor in the other direction, how to solve this image image

mkazhdan commented 1 year ago

I'm afraid that I can tell by looking at the reconstructed images. It could be the case that you are missing parts of the scene (e.g. the backs of building) so the reconstruction fits a smooth interpolant there. (If that's the case and you don't want that, you can use the SurfaceTrimmer application to remove the coarse geometry.) Or it could be the case that you have points there but the orientation of the normals is reversed from what it should be. (This happens for some applications that attempt to assign normals to unoriented points in a point cloud.)

If you would be willing to share the data, I could take a look.

glt-mhs commented 1 year ago

I'm afraid that I can tell by looking at the reconstructed images. It could be the case that you are missing parts of the scene (e.g. the backs of building) so the reconstruction fits a smooth interpolant there. (If that's the case and you don't want that, you can use the SurfaceTrimmer application to remove the coarse geometry.) Or it could be the case that you have points there but the orientation of the normals is reversed from what it should be. (This happens for some applications that attempt to assign normals to unoriented points in a point cloud.)

If you would be willing to share the data, I could take a look.

I think it's a normal problem, because I only have problems with models trained using nerfstudio, but the rest of the data works fine, and using poisson first and then ssd is much better