Evaluate dynamic handling of pig removal and reintroduction

Evaluate whether BMCTrack-d can automatically detect the permanent removal of a pig from a pen, remove the corresponding back mark from its set of known identities, and subsequently reactivate that identity if the pig is reintroduced.

Background

BMCTrack-d is designed for closed-set tracking: the possible pig identities and their back marks are known before inference. The paper explains that the system cannot currently accommodate the introduction of new back marks without retraining, while removal and later reintroduction could in principle be inferred from changes in the mean number of detections and the frequency of predictions for the affected back mark.

Although the paper outlines a possible automatic procedure for detecting removal, ignoring the associated identity, and reactivating it after reintroduction, it does not empirically evaluate this procedure. The unresolved issue is therefore whether the proposed detection-and-reconfiguration strategy works reliably in practice.

References

Thus, this situation could be handled automatically, by 1) detecting the removal via a drop in the mean number of detections, and 2) removing the affected back mark from the list of known back marks (i.e., ignoring its prediction). If the pig is reintroduced to the pen at a later point in time, the process can be reversed, i.e., detecting an increase in the mean number of detections and reactivating the back mark in the list. A practical evaluation of this is left for future work.

— BMCTrack-d: Pig re-identification and tracking via back marks in challenging camera settings  (2609.03463 - Brunner et al., 3 Sep 2026) in Section 4.4.2, “Closed-set settings”