Motion is read out by pools of direction-selective neurons in visual cortex (area MT). Watching a spiral that contracts drives the "inward" pool hard, and its gain fatigues. When the spiral stops, both pools fire only spontaneously, but the fatigued one is weaker, so the balance tips towards the opposite direction: a perfectly static picture seems to expand. This is the waterfall illusion.
da_P/dt = (1 - a_P) / tau_a adapted pool, while the spiral moves
da_N/dt = (0 - a_N) / tau_a opposite pool stays fresh
test phase: da/dt = -a / tau_r (tau_r = 5 s)
rate_P = 1.0 (1 - a_P) while adapting, 0.35 (1 - a_P) in the test
net = rate_N - rate_P = 0.35 (a_P - a_N) -> perceived direction
stimulus = sin(9 * 2 pi * r / R - arms * theta + phi)
- Same model as the other half of the pair - one shared core.js: both the 2D page and the 3D lab run these equations, so the same sliders give the same numbers.
- Turntable = phase - turning the disc by an angle alpha shifts the pattern phase by arms x alpha; check 4 verifies this exactly, which is why the spinning 3D disc is the same stimulus as the 2D one.
- tau_a = max(1.5, 6 / fatigue) s - a stronger fatigue rate adapts faster and gives a bigger aftereffect after the same stare. Recovery is fixed (5 s), so the aftereffect fades with half-life 5 ln 2 = 3.47 s (check 2).
- Honest simplifications - two lumped pools instead of millions of neurons, first-order fatigue only, no divisive normalisation or cortical feedback, spontaneous rate 0.35 of the driven rate is illustrative. Real aftereffects last from seconds to minutes depending on adaptation time.
- Correction - the previous version of this page labelled the rotation direction the wrong way round; here "contracting" means the rings really move toward the centre.