Tracking deformable manipulation

Thanks for your great work!
I found when using TAPnext++ to track deformable manipulation video, especially a long video, it performs not well.
For example, when folding a towel and then wipe the table, a lot of points on the towel are tracked to wrong positions.
If you have any insight on this, please let me know

Start with TAPnext++ and reproduce the reported failure on a long deformable-manipulation video involving folding a towel and wiping a table. Compare the tracked point positions through those actions; the issue does not identify files, tests, or a defined fix, so done is not yet specified beyond understanding or resolving the tracking failure.

Written by the indexing model from the issue text.