Fig 1.
The visual stimuli consisted of two light gray bars moving in opposite directions at constant speed (see also S2 Movie). The moving bars were embedded in dynamic visual noise, and on half of the blocks, a white noise click was presented at the time of the crossing of the moving bars.
Fig 2.
A. Reverse correlation analyses. Noisy stimuli were classified according to participants’ responses, and classification images were calculated from the mean (i.e., luminance) and the mean squared error (i.e., contrast) of each stimulus plus noise sample (see Methods). B. Luminance and contrast kernels for the aggregate observer. Warm colors represent samples positively associated to bounce responses, whereas cold colors represent samples negatively associated to a bounce response. C. Non-visual factors influencing participants’ responses for the aggregate observer. Errorbars represent the 99% confidence interval. See S1 Fig for individual observers’ data.
Fig 3.
A. Average motion energy matrix calculated from the noisy ambiguous displays (note that this is not a classification image). The plots above and to the right of the motion energy matrix represent the motion energy profile averaged over space and time, respectively. Note the drop in motion energy profiles at the intersection of the trajectories. The darkness of the lines represents the amount of total motion energy in the display (darker = more energy). To derive these plots we binned the displays in 5 groups depending on their total motion energy. The drop in motion energy, that is the difference between the maximum and the minimum motion energy of each noisy display, is linearly related in both space (B; see also A, top plot) and time (C; see also A, right plot) to the total motion energy–and hence to the contrast–of the display.
Fig 4.
Perceptual classification model.
A. Model. Visual motion energy is first computed from the stimuli through motion energy filters, and the result is linearly integrated with the auditory information and recent perceptual memory into a single estimate to determine the Z-score of streaming/bouncing responses. B. Scatterplot of empirical vs. predicted responses for the aggregate observer. Each dot is the average of 608 responses. Light red area represents the 99% confidence interval of the identity line. C. Luminance and contrast kernels calculated from the model responses. D. Cross-correlation between predicted and empirical kernels. The red lines represent the thresholds for statistical significance (p = 0.05) as calculated based on the permutation test.
Fig 5.
A. Luminance and contrast kernels estimated from the motion energy model for light (top) and dark (bottom) moving bars. B. Results of Experiment 2. The bars represent the probability of responding bounce for high (HI) and low (LO) motion energy drop and for light and dark bars. Errorbars represent the standard error of the mean. C. Scatterplot and bagplot of the probability of responding bounce for stimuli with high and low motion energy drop. Thin dashed lines connect data from the same participants in the two lightness conditions. The red cross represents the depth median.