Replicating and Extending Published Research using Meta Quest 3 V4 Headsets

Figures 1 and 5 from Stoner and Blanc (2010). Figure 5 shows both conventional depiction of delayed-onset design and the feature-trajectory depiction that allows easier appreciation of both spatiotemporal dot continuity and feature history. We initally wanted to replicate Stoner and Blanc's findings for the standard (i.e. no feature swaps) conditions and the motion/color swap conditions. Figure 5 A and C show the no motion/color swap conditions and Figure F and H show the motion/color swap conditions.

Live stimulus demos — motion at ½ speed, translation duration 4× actual (176 ms vs 44 ms). Red field is always-on; green field has delayed onset (750 ms) and serves as the attention cue. On cued trials the green (delayed-onset) field translates; on uncued trials the red (always-on) field translates.

Figure 4. Inputs feeding the biased-competition circuit. (A) The directional time-varying input into the model for a cued, no-swap trial (see Figure 3). (B) The three direction-selective model neurons of "Stage 1" receive this excitatory directional input. We can think of these three neurons as belonging to a single MT hypercolumn (see Figure 2). These three neurons each have a directional preference corresponding to one of the three predominant directions within the collective receptive field of that MT hypercolumn: Down, Right, and Up. The responses of these three Stage 1 neurons adapt over time, mediated by recurrent inhibition from the I unit and modeled by the equations shown. These adapted responses then feed into the Stage 2 translation detector R₍TD₎, with R₍T₎ providing excitatory (W⁺) input and R₁ and R₂ providing inhibitory (W⁻) shunting input. Taken together, these interactions implement the biased-competition computation of Reynolds et al. (1999). In this cued example, Down appears first and is therefore more adapted by translation onset, while Up arrives later. Hence the translation competes with a weaker inhibitory input than occurs with the uncued condition. In our simple model, we take the output of this translation detector as our measure of how well the model detects the translation (see Figure 5).

The stimulus in feature coordinates — replication of Stoner & Blanc (2010), Figs. 1B & 4. Each dot field is a trajectory through three motion tracks (clockwise rotation, translation, counter-clockwise rotation). Following S&B's convention, line style = field identity (dotted = the test field that translates; solid = the competitor), line colour = dot colour, and vertical position = type of motion. The same (dotted) test field translates in every condition — only the onset (which field is delayed) and the swap differ. No swap (left column): the test field briefly notches into TRANS and returns to its rotation. Motion swap (right column): at translation the two fields exchange rotation directions — the test field keeps its identity (still dotted) but ends on the opposite track. The consequence is decisive: after a swap, a cued-swap trial's motion content matches an uncued-no-swap trial, forcing any direction-only model to predict that the cueing advantage reverses.

Directional inputs S(t) — what actually drives the model, across all four trial types. The feature trajectories re-expressed as the binary drive to each motion channel (rotation drive = 50, translation = 25; Mode 1). The test field (dotted) drives its current rotation track plus the brief TRANS pulse, with a gap where its rotation is interrupted by the translation; the competitor (solid) drives the other rotation track. Under a motion swap the test field keeps its identity (dotted) but its drive shifts to the opposite rotation track. These per-channel drives are the model's complete input — and the stage at which the two surfaces' identities collapse into direction-only signals.

Model responses across all four trial types. Top row: the two rotation channels — dotted marks the channel interrupted by translation (it shows the brief gap), solid is the other. Middle row: the detector's two inputs — the summed normalization pool I = R₁ + R₂ (black) and the excitatory translation drive E (blue, dotted). Bottom row: the translation-detector response R₍TD₎ with its peak. Under no swap the cued field gives the larger response (61 vs 46 — a +33% advantage). Under motion swap the cued-swap and uncued-swap responses are identical to the opposite no-swap conditions, so the advantage reverses (46 vs 61, −25%). The behavioral data show no such reversal — the model's clean falsification, and the motivation for surface-identity selection.

Colour swap — and why the model is colour-blind. Colour (the dots' red/green) is carried by line colour; field identity by line style. A colour swap flips the line colour at translation while the line style — the identity — is untouched, and it can be combined with a motion swap or applied on its own, fully independently. Our current model has no colour input: it is deliberately colour-blind. That is acceptable because colour differences between the surfaces have been shown not to be necessary for the surface-selection effect — the cueing advantage survives when the two fields share the same colour. Colour is shown here only to make the identity-versus-feature dissociation explicit (line style stays fixed while colour and motion vary); it changes none of the model's predictions.