Confusion Lines

Their eyes uses the Machado, Oliveira and Fernandes (2009) model. Severity 0 to 100 tracks how far the weak cone's sensitivity has shifted toward its neighbor (up to about 20 nm, where the person becomes a dichromat). Each pixel loses the one color component the profile cannot sense; everything else stays. That is why purple and blue merge for a strong protan, and red, orange, yellow and green crowd together for red-green types.

A simulation is right when the person it describes cannot tell it from the original. Side by side (in More) is the check.

Assist picks its method from severity. Below 90 it enhances: the lost component is boosted along a direction that leaves brightness unchanged, small differences get more gain than large ones, local red-green texture is sharpened, and nothing clips. Strength starts at s/(1−s) for severity s, so a 60 profile gets about 1.5x. At 90 and above it recolors instead (Fidaner, Lin and Ozguven daltonization), moving lost information into channels a dichromat still has, since there is nothing left to boost.

Assist looks overdone to normal eyes, because it is built for eyes that are weak on one axis. "Show Assist through their eyes" in More fixes that, so you can judge what they would gain.

Live camera applies every view to each frame and keeps the last 30 seconds in memory for Replay. Pulse amplifies tiny skin color changes from blood flow and estimates heart rate from the center of the frame (a demo, not a medical reading). Heat palette is for a plugged-in thermal or NoIR camera; phone cameras block infrared.

The test uses generated plates whose figures differ only along one confusion line, a D-15 style arrangement scored with the Vingrys and King-Smith (1988) method, and a matching slider that finds the point where the simulation stops being invisible to you. Screens vary, so this is screening, not a diagnosis.