Convert images to RGB in examples

[taylor-arnold]

As mentioned in #36, and tested on my own data, the input to MTCNN should
be an image in RGB format. The OpenCV library load images as BGR and needs
to be converted.

Note that using the BGR format works okay for high-quality faces, as in the test
example image, but fails to detect more difficult (small, grainy, profile, ect.) faces.

[ipazc]

Note that steps_threshold = [0.6, 0.7, 0.7] can be adjusted in the MTCNN constructor to increase detection rate depending upon the color format. Even though it is true that RGB color format reports better performance with the default thresholds.