SAM3 longer vidoe sequences

Hi,

I am trying to develop auto-labeling with vidoes, being about 9000 frames of sequences.
I can't get the streaming working properly, thus my memory get full to quickly.
Further, I am trying to track multiple objects, "person,car,truck".

Does anyone have some good material I can read to or hints on how to tackle this?

The issue names no file, test, or entry point. Start with the repository's example notebooks and investigate the streaming and multi-object-tracking paths described; done means processing roughly 9,000 frames without memory exhaustion while tracking people, cars, and trucks.

Written by the indexing model from the issue text.