Fig. 3: The performance of human observers agrees with the new dim light detection model where performance is fundamentally limited by retinal thresholding nonlinearities along the ON pathway.
From: Primate retina trades single-photon detection for high-fidelity contrast encoding
a The psychophysical experiment. In the detection task, the observer had to report which of the two intervals contained the flash. In the discrimination task, the observer had to report which of the two intervals contained the brighter flash. b Psychometric functions of a representative observer: the detection task (blue) and the discrimination task (red) for one reference flash intensity. c ΔIJND (75% correct) as a function of the reference flash intensity (Dark for detection) for all observers. Thin lines correspond to individual observers, the dotted line denotes the example observer shown in (b), and the thick line denotes the population average. d Average ΔIJND (mean ± SEM) as a function of the reference flash intensity for human observers (orange, n = 5 observers), OFF RGCs (magenta, n = 11 cells), and ON RGCs (green, n = 54 cells). e Model predictions (thick dashed lines) for ON RGCs (M1, number of subunits: n = 3, θ1 = 2; similar to the parameters used in Ala-Laurila and Rieke10), an ON RGC model with twice the number of subunits (M1, n = 6, θ1 = 2), and for a model with two thresholding nonlinearities (M2, n = 6, θ1 = 2, θ2 = 2): one in the retina and one downstream of the retina. Same average ΔIJND (mean ± SEM) for a human observer (orange) and ON RGCs (green) as in (d). The dotted black line shows the theoretical performance limit set by quantum fluctuations in the number of stimulus photons. For the model robustness analysis and parametrization, see Supplementary Figs. 3 & 4. Source data are provided as a Source Data file.
