[Work Log] Visualizing/Debugging BG ML

June 12, 2013
Project Tulips
Subproject Data Association v2
Working path projects/​tulips/​trunk/​src/​matlab/​data_association_2
Unless otherwise noted, all filesystem paths are relative to the "Working path" named above.


We now have a new algorithm for computing background curve Marginal Likelihood. Lowering noise sigma \(\sigma_n\) should rule out bad matches.

Task: Re-run background candidate matching with new algorithm and roughly-trained \( \sigma_n \).


If the threshold is set right, the results are improved, but we still have some false-positives and false negatives.

It's still unclear whether we can get good results without thresholding, since we haven't computed the noise ML using the new algorithm, so absolute numbers are meaningless.


smoothing_variance_2d: 0.2500
    noise_variance_2d: 10
     position_mean_2d: [2x1 double]
 position_variance_2d: 1.3629e+04
     rate_variance_2d: 0.4962
   smoothing_variance: 1.0000e-04
       noise_variance: 10
        position_mean: [3x1 double]
    position_variance: 62500
        rate_variance: 2.2500
      smoothing_sigma: 0.2000
       nlise_variance: 10
    noise_variance_bg: 0.1038


data = offline_pair_candidates(data, params, 0, 1, 1, 'bg');
cands = tmp_get_cands(data);
visualize_bg_cands(data, cands, 250)


Matched curves appear in white, unmatched appear in gray

False negatives

bad matches

False positives:

bad matches

Next Steps

Posted by Kyle Simek
blog comments powered by Disqus