Why we need a coarse-to-fine strategy to solve the optical flow problem (feature tracking) in practice? If we do not use such kind of methods, what will happen?
If an image has larger motions, then the traditional Optical Flow methods like the non-iterative Lucas Kanade can perform badly because they assume motion is small (subpixel). The coarse-to-fine method helps a lot because in a coarser resolution you can assume that the brightness constancy can hold.