Abstract
Humans perceive the angle separation (AS) between two overlapping random-dot stimuli that move transparently in different directions to be wider than it actually is when the veridical AS is less than 90°. The neural basis for this illusion of direction repulsion remains unclear. We recorded from neurons in middle-temporal (MT) cortex of two fixating macaques. Visual stimuli were overlapping random-dot patches (diameter 7.5°) moving in two different directions. The AS was 30°, 45°, 60°, 90°, or 135°. We varied the vector-averaged direction of the bi-directional stimuli to characterize a neuron’s response tuning curve, and also measured each neuron’s direction tuning to a single patch. Visual stimuli had eccentricities from 1.5° to 29° (median=6.3°). We fitted tuning curves to bi-directional stimuli as a weighted sum of the responses to the two component directions, plus a multiplicative term between component responses. In the model fit, we allowed the “component directions” to deviate from the veridical component directions, with either a wider or narrower AS. The best-fit component directions averaged across neurons (n≥96) had mean AS of 54°, 69°, 94°, and 133° for veridical AS of 45°, 60°, 90°, and 135°, respectively. Consistent with human perception, we found a significant effect of direction repulsion at AS of 45° and 60° (p<0.0001), but not at 90° and 135°. Furthermore, the ratio between the overestimated AS and AS (i.e. ∆AS/AS) decreased from 0.2 to -0.015 as the AS increased from 45° to 135°. However, in a smaller neuron sample (N=38), we did not find direction repulsion at AS of 30°, which may be due to stimulus eccentricities and the small AS. In conclusion, we found that within a range of AS, MT neurons encode transparently moving stimuli as if they have a wider AS, which would allow a decoder to read out repelled motion directions.