A growing body of evidence suggests that optic flow is processed by specialised 2-D motion mechanisms. We asked whether the visual system has parallel, or rapid serial, access to representations of optic flow components in a spatial 4AFC task. Random-dot kinematograms (144 dots per interval) depicting expansion, rotation, deformation (horizontal shear+vertical shear) or one of these components summed with translation, were presented in four spatially abutting circular windows (2.65 deg in diameter), and were temporally modulated by half a cycle of a 300 ms cosine-wave. Within a session, stimuli were of the same type, but the target had opposite sign and was selected with a mouse and cursor with feedback. Systematic local cues were removed by randomising (i) the orientation of the windowing configuration, and (ii) the dot speeds between intervals (speed gradient varied between 2.4% and 6%; translation varied between 16 and 40 min arc s−1). Preliminary results (average SE=3.4%) showed that in the absence of translation, performance was close to chance (25% correct) for rotation (23% correct) and deformation (27% correct), but was good for expansion (61% correct). The addition of translation had no effect on rotation but improved deformation (58% correct) and impaired expansion (22% correct). In experiment 2, unlimited stimulus repetitions were allowed and performance improved (>93%) for all conditions, though, as predicted from experiment 1, reaction times were fastest for expansion and deformation-plus-translation. Importantly, only these two conditions produced unambiguous 3-D perceptions of the stimuli, suggesting that surface slant and motion in depth are coded by mechanisms more rapidly accessible than those subserving general extraction of 2-D motion.
Spike Burst Coding of Translatory Optic Flow and Depth from Motion in the Fly Visual System