fix: restore from Gibbs indices to measurement indices #3
Conversation
|
Hi @DoHaiSon, thanks for submitting this patch. I understand that this measurement index issue arises when gating is enabled. In my implementation, while drafting the paper, enabling gating led to a decline in tracking performance. So I set the default gating is OFF and did not report any result using gating in the paper. Could you share the tracking performance results after applying this patch? It would be greatly appreciated if you could include this in a comprehensive document, outlining the performance metrics. I’d like to confirm that performance remains stable and that there are no unintended side effects before merging your request. |
|
The OSPA^2 and MOT metrics below are from my run on CMC_1 with feat turned off and multiple hyp. 
|
2 participants
This pull request introduces logic to restore measurement indices from internal Gibbs sampling indices in both the C++ and Python multi-sensor GLMB-UKF filter implementations. This ensures that after internal processing, the indices used for measurements correctly map back to the original measurement indices, improving consistency and correctness in downstream processing.
Restoration of measurement indices:
cpp_ms_glmb_ukf/src/run_filter.hpp, added a loop to convert Gibbs indices back to measurement indices for all relevant entries inupdate_hypcmp_tmpusing themindicesmapping after hypothesis assignment.ms_glmb_ukf/run_filter.py, inserted corresponding logic to restore measurement indices inupdate_hypcmp_tmpfor all non-off (active) tracks using themindicesmapping.