Will there be an accumulation of errors in the predicted camera poses?

[sIHURs]

I have 210 input images, and the camera poses are distributed on a sphere, as shown in the figure — the right side of the image corresponds to the bottom of the sphere. The camera poses should roughly form a circular pattern (as seen on the upper part of the sphere on the left side, where the predicted poses are displayed).

When I split the 210 images into two halves and predict them separately, the predicted poses for the bottom part of the sphere turn out much better.

top:

bottom:

What could be the reason for this? Thanks a lot! :D

Yifan

[jytime]

Hi I cannot see the input images, while a wild guess is, probably this results from a kind of doppelgangers?

https://doppelgangers25.github.io/doppelgangers_plusplus/