Question

I’m running a HaarTraining process using OpenCV 2.2 (compiled with OpenMP). The training progress shows:

+----+----+-+---------+---------+---------+---------+
|7468| 44%|-|-2.712450| 1.000000| 0.719459| 0.108393|
+----+----+-+---------+---------+---------+---------+
|7469| 44%|-|-2.712449| 1.000000| 0.719459| 0.108393|
+----+----+-+---------+---------+---------+---------+
.......

What does the number 7469 mean (I only have 7000 samples)? How can I know when the process is going to finish?

Thanks!

Answer 1

N means the current feature for this cascade. (see http://cv-kolaric.blogspot.com/2008/03/output-from-haartrainingexe.html)

+----+----+-+---------+---------+---------+---------+
|  N |%%SMP|F|  ST.THR |    HR   |    FA   | EXP. ERR|
+----+----+-+---------+---------+---------+---------+
N = current feature for this cascade (seq->total),
%%SMP = percentage of samples used, if trimmings enabled (v_wt)
F = '+' if isFilpped, if symmetry is specified (v_flipped), '-' otherwise
ST.THR = stage threshold, 
HR = Hit Rate based on Stage threshold (v_hitrate / numpos), 
FA = False alarm based on Stage threshold (v_falsealarm / numneg)
EXP.ERR = Strong classification error of adaboost algorithm,
based on threshold=0 (v_experr)

Different features are calculated for all your samples. E.g. x-y derratives, contrast-ratios etc. etc. Depending on the difference between your samples, more features are calculated, and it seems your positive images are just too different from each other, to categorize them.

Try to pre-process your images first, before training. E.g. remove background, find edges, threshold the images, flood-fill. Anything that can make your positive image set look more alike