Single fly psychophysics
Here is a tethered fly walking on an air-suspended plastic ball. In an experiment, the fly would be surrounded by 3 screens that display interesting visual stimuli. By measuring the ball’s rotation, we can infer the fly’s walking behavior in response to a given stimulus. This gives us a tool for measuring the fly’s perception of all sorts of interesting stimuli.
The following 2 movies demonstrate the reverse-phi visual illusion, which both flies and humans are susceptible to. The movies consist of white and black dots on a gray background, updated each frame. In the first movie, dots are placed randomly in one frame, and in the subsequent frame, they are moved 5 pixels to the right before disappearing. This produces a percept of motion to the right, as one might expect. The second movie demonstrates the “reverse-phi” illusion. This time, white and dark dots are placed randomly in one frame, and in the subsequent frame, they are moved 5 pixels to the right before disappearing, as before. The difference is that this time, their contrast is inverted when they are moved to the right, so that white –> black and black –> white. With this contrast-inverting update rule, humans (and flies!) perceive motion to the left, even though the dots are moving to the right. This illusion is consistent with models of motion estimation in vertebrates and invertebrates. Cool!
(Phi stimulus, tuned for human consumption)
(Reverse-phi stimulus, tuned for human consumption)
Correlations with different intervals
To quantify how flies perceive different correlations, we have developed specialized stimuli that present positive or negative correlations, as above, but at different specific spatiotemporal offsets. These are visible in the movie below. In it, all displacements/correlations are to the right. The movie first shows positive and negative correlations with a 1-frame interval, which are perceived as strong right and leftward motion. It then shows positive and negative correlations with a 2-frame interval. More details are available in Salazar-Gatzimas et al., Neuron, 2016.
We use other interesting correlations to probe the visual systems of flies (and humans). Following a beautiful method devised by Jonathan Victor (Cornell), we created binary stimuli containing correlations between triplets of points in time and space. Classical models of motion estimation rely on correlations between pairs of points, and those models predict no motion would be perceived in three-point correlation stimuli. In fact, as Victor first showed, humans perceive weak but consistent motion in such stimuli. We’ve created an example below, for your viewing pleasure. Although the percept is weak, if we put you in a dark room and forced you to choose which direction each these patterns is moving (on average), you would choose the same direction for each pattern about 90% of the time, and you would largely agree with everyone about the direction of motion.