Most of the times I see people using the RGB channel for the training set. While it is good for facial detection but I have not found it as effective as I thought it would be! So, should I try converting the image into a different (possibly a combination of different color transitions) color space to train an object identifier?
I cannot afford to do find it out experimentally as there is a higher cost associated with that(AWS).
P.S. I am using dlib's dnn_mmod_ex.cpp!