Dear Author:
This is a very valuable work, which is significant in some practical situations.
I did some validation on it, when I trained the point cloud and pose of a scene, I took out one of the images with very high confidence and tested it, firstly generated its depth map and backprojected it as a point cloud, after that, after projecting it to the global coordinate system using the camera pose obtained from the ACE0 computation, I found that its pose is wrong, but it is not clear what caused the error, please advise me.