Invisible Corners
Try a few cycles of the animation. As the red squares turn into thin L-shapes, you should experience a switch from coherent motion (a square going up and to the right) to incoherent motion (pairs of segments moving vertically and horizontally in counterphase). McDermott, Weiss & Adelson (1999 ) suggested that the change is due to a higher-order constraint having to do with connectivity in depth of contours. During motion integration, this higher constraint overcomes local t-junctions that suggest occlusion. In other words, when the L-shapes are not thick enough to hide the corners of the square, but you don't see those corners, the overall occlusion interpretation is weakened and so is its effect on the integration process. As most higher-order effects, the effect of invisible corners tends to be unstable. You should fixate the center of the square and try to see it as a single, homogeneous surface.

created by Fauzia Moscaand Nicola Bruno