orientation correction technique about ab3dmot

[zouxy09]

I think you miss some considerations in the orientation correction technique, and I improve it and get the following result:
=================evaluation with confidence threshold 3.398000==================
MOTA MOTP MODA MODP MT ML IDS FRAG F1 Prec Recall FAR TP FP FN
0.7657 0.7916 0.7657 0.8375 0.7004 0.0727 0 50 0.8979 0.9327 0.8656 0.2229 24793 1790 3850
=====================evaluation: average at all thresholds======================
AMOTA AMOTP APrec
0.3948 0.7467 0.8444
All the items are little better than the results in your paper. I argue that your orientation correction technique is not suitable to all scenes, and the limit of kitti dataset cover up the mistakes.

[xinshuoweng]

Thanks for pointing out! I believe you are right. This is a bug I have missed.

I have pushed a commit to fix this bug and got exact same results as yours.

Thank you!