You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Yes, this describes a bounding box that encompasses the whole scene. The grid of networks will only live in that bounding box. The size of the bounding box could be automatically inferred with the SfM algorithm (e.g. COLMAP) that you have to run anyhow for estimating the camera poses of the training views. SfM outputs a point cloud. The size of the bounding box must be such that it contains the entire point cloud.
Analogically, you have to adjust the "near" and "far" hyperparameters such that the entire scene is sampled. This can also be done based on the sparse point cloud from SfM.
Any suggestions on how to estimate the global domain size?
The text was updated successfully, but these errors were encountered: