If anyone's interested there's a neat "Milia Blocker" flash tool to help get your reaction times up.
http://labzerogames.com/iheartteyah/MilliaBlocker_v0.3.swf
Only press a button for the overheads.
ugh, screw this
I keep getting faults, because my natural reaction to HairCar is to Downback,Back,Downback+HeavySlash. Verifying for not-HairCar is really slowing me down. And BadMoon is missing the smoke-puff behind her, for when she jumps.
So how many people actually look directly at their opponent's character? I always found that to be too much visual information to parse. Animation changed, what is it? Walking. Animation changed, what is it? Whiffing a jab. Animation changed, what is it? Man, what an ugly color. Oh, fuck, overhead!
Instead of focusing directly on my opponent's character, I look at the space between us. And use peripheral vision to look around my opponent's character. For example, against Millia: If her sprite expands behind her, then it's HairCar or BadMoon. If her sprite contracts inward and moves upward, then it's her standing overhead. If she contracts inward and moves downward, then it's Dust. Simplifies things for me.
Basicly, if character enters certain areas around themselves, or leaves already occupied areas, then it can only be a certain move. Or if it can potentially be many moves, you can usually order them on a timeline. And then order your inputs to match up with the possibilities. Like VF fuzzy-guarding.
Characters like SkullGirl's Double drive me insane, because her sprite can be anything from huge monster to tiny teacup. And then, 3D fighters, too much tweening.