3D Consistency of RealEstate10K

[1ssb]

I have been trying to get a point cloud to check for the 3D consistency of the scenes developed in Real Estate sampling. It doesn't seem like they are 3D consistent for the experiments I performed. Could it be that there is some form of anisotropic scaling which might be happening to the pixel Nerf coordinate to world coordinate. Where could I be going wrong?

Note: I am trying alternative approaches.

[edwinseychelles]

Same here, did you find what could be going wrong? There does not seem to be any sources of anisotropic scaling.