Fig 1.
(a) Optic flow components on the retina of a mobile observer. The retinal motion (blue, left panel) is the sum of motion created through self-motion relative to world-fixed stationary environment (center panel) and the motion created by objects that move independently from the observer (red, right panel). The visual system could recover the world-relative motion of objects (red arrow, left and right panels), by subtracting the self-motion component (center panel) from the retinal pattern (left panel). (b) Neural algorithm proposed by Layton & Fajen [4] to recover world-relative object motion. The call-out on the right is a polar plot showing direction responses to the moving object. MSTd cells that respond to the observer’s self-motion send feedback to suppress MT cells (light blue region, right panel) that signal the retinal motion (light orange region, right panel) consistent with the preferred MSTd tuning (open arrows, left panel). Suppression shifts the direction signaled by the MT population toward the world-relative direction (dark orange, right panel).
Fig 2.
(a) The model contains two major pathways: MT+/MSTd subserves self-motion (top) and MT−/MSTv subserves object motion (bottom). The model is dynamic and estimates self-motion and object motion simultaneously. Cells in MSTd model physiologically supported populations tuned to full-field radial patterns, speed (band-pass, speed summating, gradient), and non-direction dependent disparities. Cells in MT−/MSTv have RFs with inhibitory surrounds due to local and feedback mediated suppression. Feedback comes from dominant MSTd units and suppresses MT−/MSTv units locally tuned to the direction, speed, and disparity that match the estimated self-motion motion pattern. This shifts population object motion responses from a retinal to world-relative reference frame. (b-d) Model MT tuning curves for direction (b), speed (c), and disparity (d). *MSTd units tuned to ground patterns were added in Simulation 2. **Disparity tuning was incorporated in Simulation 3.
Fig 3.
Overview of suppressive feedback mechanism.
MT− and MSTv units are inhibited depending on how their tuning properties relate to the corresponding local subregion of the most active MSTd unit’s RF. (a) The pattern selectivity of the MSTd unit is indicated on the top-right and the RF of some MT units are shown in left superimposed circle. Highest peak curve in plot shows that MT suppression is greatest when the MT direction tuning locally matches (leftward; 0° mismatch) the MSTd RF; units with mismatching direction tuning receive progressively less inhibition. Other curves show that inhibition drops off as speed tuning locally differs from the MSTd RF. (b) MT− and MSTv units receive the most inhibition when they match the congruent speed-disparity tuning of the MSTd unit (circled diagonal). For example, among units with RFs near the FOE, those tuned to slow and far receive more inhibition than those tuned to slow and near.
Table 1.
Model unit counts within each area.
*MSTd units tuned to ground patterns were added in Simulation 2. **Disparity tuning was incorporated in Simulation 3.
Fig 4.
(a) Fronotoparallel displays used in Simulation 1. Optic flow depicting simulated self-motion at different rates of approach toward a frontoparallel plane of dots. The red arrow depicts the retinal motion of a small moving object. We scaled the length of the object vector 3x for visibility. (b) Model MSTd gradient cell RF subunits corresponding to sensitivity to the five different model MT peak speeds, which span the dynamic range across all frontoparallel stimuli (i.e. not necessarily any particular display). Increasing gradient cells prefer an increasing speed gradient and have the region of maximum sensitivity to the slowest (fastest) speed near the center (periphery) of the RF. The speed gradients in (a) stimulate a subset of the gradient cell annular subunit regions. Connection weights are depicted before normalization.
Fig 5.
(a) The accuracy of human object motion judgments in Simulation 1 (frontoparallel plane), as measured by flow parsing gain (reproduced from Niehorster, 2013), compared with model object direction estimates. (b) Human judgments and model object direction estimates compared, expressed as angles. 0° indicates the retinal object direction and positive angles indicate shifts toward the world-relative direction.
Fig 6.
(a) Object motion signals in MT− during fronotoparallel plane simulation. Blue is the object motion signal in MT−, green is the inhibitory feedback signal from MSTd, red is the surround (local) inhibition within MT−. y-axis is activation, x-axis is direction (0-360 deg in 15 deg increments). (b) Shift in the MSTv activity distribution over time (MT− demonstrates a qualitatively similar evolution). (c) Time course of object motion direction represented by area MSTv units. A zero shift value indicates object motion in a observer-relative reference; positive values indicate shifts in the world-relative direction. Self-motion speed is 1 m/sec and object motion speed is 6°/s. (d) Activation of the most active gradient (gray) and speed summating (black) MSTd cells for the different self-motion speeds from the Experiment 1 frontoparallel plane stimuli. (e) MSTd band-pass cell activity that reflects an estimate of observer’s self-motion direction. The x and y axes correspond to the spatial coordinates of the optic flow. The overall activation is weak and variance in the distribution is high, indicating uncertainty in the self-motion estimate.
Fig 7.
(a) Optic flow from the ground plane scenario used in Simulation 2. The red arrow shows the observer-relative motion of a small moving object. We scaled the length of the object vector 3x for visibility. (b) Model and human flow parsing gains. Error bars denote SEM.
Fig 8.
(a) Regions of a ground unit’s RF tuned to a central heading that exhibit maximal sensitivity to the indicated MT+ speeds and directions. (b) Connection weights among the different speed-tuned inputs along a subset of the directions for the slowest speed subunit. Weights shown prior to normalization.
Fig 9.
Model results for ground plane scenario (Simulation 2).
(a) MSTd band-pass cell activity. (b) MSTd ground unit activity. (c) Excitatory and inhibitory signals integrated by MT units that process object motion. (d) Time course of the shift in the object direction signal represented in the model MSTv toward the world-relative direction. Black curve (frontoparallel plane) reproduced from Fig 6c. Blue curve corresponds to simulation of 4° object ground plane condition.
Fig 10.
(a) Full, (b) No Local Depth, and (c) No Local Frontal View conditions (Simulation 3). Reproduced from Niehorster & Li [2]. (d) The flow parsing gain achieved for these conditions by the model and the human observers. Error bars denote SEM.
Fig 11.
(a) MSTd band-pass cell activity for the Full Stereo condition. (b) Kurtosis of the most active MSTd subpopulation response, which measures the peakedness of distribution, in each simulation. (c) Strength of the recurrent signal generated by the most active MSTd subpopulation in each simulation (Eq. S21). A zero value in this panel indicates that the MSTd signal did reach the appropriate threshold to engage the recurrent mechanism. (d) Time course of the object direction signal in model MSTv. Black and blue curves reproduced from Fig 9d.
Fig 12.
Object motion signals in MT in the (a) Full Stereo, (b) No Local Depth, and (c) No Local Frontal View conditions.
Blue is the object motion signal in MT−, green is the inhibitory feedback signal from MSTd, red is the surround (local) inhibition within MT. y-axis is activation, x-axis is direction (0–360° in 15° increments). Local inhibition refers to the inhibitory signal due to feedforward lateral inhibition from units tuned to similar directions, speeds, and disparities in MT Layer 4/6.
Fig 13.
The flow parsing gain achieved by the model and the human observers for different self-motion speeds with stereo optic flow (Simulation 3).
Error bars denote SEM.