Facial landmarks detection using two ways, one with large model which has better accuracy, and another which is lightweight and fit weak devices
Classification \
Simply single input(single object in the image) --> single output \
It should be capable of saying that this image has a cat with probability of xx.yy%, but can't say where.
Detection \
Single image(Containing multiple objects)--> outputs where are these objects and what they are.
Localizatoin \ By Localization we mean how put rounding boxes around the object.
Slide a window over the whole image.
Notice the scale effect. It will not be able to detect face from small window. \
To make less computations you may use stride, You don't have go pixel by pixel.\