How to choose the method for SImd detection

[Palle Dinesen]

First of all, I am very impressed with this work and the potential speed advantage

Question: How can you choose with method (such as 32fp or 32fi) for detection from the high level interface? Loading the classifier determines the method, Haar or LBP, but I don't understand how to choose which function is actually used?
Also in the test suite what is the difference between for example
DetectionHaarDetect32fi<0> and
DetectionHaarDetect32fi<1>?

Thanks,
Palle

[Yermalayeu Ihar]

1) Conventional notations: 32fp - 32-bit float, planar image sum(detect objects with step 1). 32fi - 32-bit float, interleaved image sum (detect objects with step 2).
Haar and LBP detector detect objects at different scales. -32fp is used in order to detect objects for upper scale levels of image pyramid (scale > 2). 32fi is used for lower scale levels of image pyramid (scale < 2).

2) This is testing of different type of Haar cascades:
DetectionHaarDetect32fi<0> - Haar cascade without tilted features, DetectionHaarDetect32fi<1> - Haar cascade with tilted features.