[Work Log] Sampling - implemeting offline pair candidates

September 26, 2013
Project Tulips
Subproject Data Association v3
Working path projects/​tulips/​trunk/​src/​matlab/​data_association_3
SVN Revision 15229
Unless otherwise noted, all filesystem paths are relative to the "Working path" named above.

Problem with revamped index-estimation code: it fails miserably on poorly-triangulatable hypotheses.

The heuristic triangulation explodes in size, probably because triangulated points are all over the map (need to confirm this).

Adding error codes; if any part of the pipeline can't complete for any reason, set error code and return. On error, the MCMC move will simply reject the proposal.

Issue with build_full_correspondence: input track is destroyed and replaced.

Refactored to replace make_correspondence with init_track.

make_correspondence -> init_track
make_trivial_correspondence -> init_trivial_corespondence

Running offline_pair_candidates_2. Seems to be running without crashing.

Takes about 10 minutes; seems fast considering all the new logic we've added (re-triangulation, 2D DTW). Probably the mex'd DTW is helping. Some profiling will likely identify some low-hanging optimization fruit here, too.

Need to inspect triangulated curves when done.

Need to see if background subtraction classifier will help prune.

Oops, spoke too soon. Crash at about 75% completion.

Random bug in candidate-proposal generator. wasn't handling the "no candidates" case, resulted in index-out-of-bounds error.

Posted by Kyle Simek
blog comments powered by Disqus