this seems to be the implementation of the 2018 Proceedings of the IEEE paper Message Passing Algorithms for Scalable Multitarget Tracking.

Similar implementation can be found at https://github.com/BaiLiping/BP_SLAM for BP SLAM, which is implemented by the TU Graz team based on this code.

It seems that the original implementation is by Jason Williams in his paper https://arxiv.org/abs/1203.2995 Marginal multi-Bernoulli filters: RFS derivation of MHT, JIPDA and association-based MeMBer. The difference between these two set of methods is that Meyer's methods, all the measurements initiated new tracks. If that measurement is generated by clutter, that "new track" would die out.

The difficulty with implementing this particular set of code comes from the loopy part. The notations are extremely confusing.

Matching the paper with this implementation might be the best way to understand things.