Traincascade and symmetric images

asked 2014-02-04 14:46:11 -0500

Bartis Áron gravatar image

I've searched, but couldn't find any reference. Has the newer "traincascade" any concern about the symmetric images? For example logos can't be mirrored, but faces yes and thus the detection could be improved. I know that about the "haartraining" that it has any, but the "traincascade"?

edit retag flag offensive close merge delete



If you mean that mirroring your training data on over the y axis to improve your detection rate, then yes traincascade will improve, especially if your head is not 100% aligned correctly, since you will have slightly different training samples. However, do use enough samples in order to avoid overfitting by this measurement.

StevenPuttemans gravatar imageStevenPuttemans ( 2014-02-05 03:03:20 -0500 )edit

Yes, I meant that, but I thought, that I can solve it automatically either with traincascade or createsamples.

Bartis Áron gravatar imageBartis Áron ( 2014-02-05 04:02:26 -0500 )edit

Nope it isn't integrated to do that automatically. Just write a small script to double your training set by flipping them all. That is what I do most of the times.

StevenPuttemans gravatar imageStevenPuttemans ( 2014-02-05 05:06:08 -0500 )edit