Using importKnownPoses results in worse data

[MasonAtor19]

Hello all! I've been trying to get the importKnownPoses to work well for me. I have a pipeline that, most of the time, runs well without it and can generate good solves for most of the cameras. However, there are some datasets that it isn't able to solve many images, so I've tried adding known poses generated from lidar data. These poses are a rough estimate and a good place to start, but can always have some inaccuracy.

When using importKnownPoses in the pipeline, and importing all the poses as SfmData, the SFM node seems to adhere too much to those poses, and the inaccuracies in the lidar poses are too apparent in the final SFM data. I've also tried only specifying only the views and intrinsics as SfmData input to importKnownPoses, and supplying the lidar poses through the node's Known Poses Data, but this results in an even worse solve.

So my question is: what is the right way to use the importKnownPoses in this scenario? These lidar poses should act as a suggestion for where to start image/feature matching, but not contrain the scene too much. In the pipeline, I have Feature Matching using known pose data, so I believe these poses are honored, but it seems like I might be misusing the import node