they can do calculations on a pixel-by-pixel basis, as opposed to standard methods which use are forced to compress information along the way through pooling/st

they can do calculations on a pixel-by-pixel basis, as opposed to standard methods which use are forced to compress information along the way through pooling/strided convolutions I'm not sure I understand your point. Why do you think dilated filters produce a better global view than pooling?