Visual Pathway User Guide
Use Auric Artisan Visual Pathway to explore how color stimuli move through cone responses, opponent channels, retina, LGN, V1, V2 and V4 proxy stages.
Overview
Visual Pathway follows one colour, or one image, through five stages of the early visual system: cone signals, opponent channels, retinal centre-surround, a V1 Gabor bank with its hypercolumn map, a V2 boundary signal and a V4 colour cell. It is an educational approximation. Every constant that is this tool’s own says so in the Data register, and nothing derived from one is reported as a published value.
1. What it does
+- Tool URL: open Visual Pathway.
- Six tabs. Lab is the workbench. Pipeline shows where a colour goes before any cortex sees it. Cells shows every stage over the same stimulus. Data is the register of what each figure rests on. Export takes a list of colours or the active map. Reference holds the standards, the formulas, the citations and the research notes.
- Everything runs on your machine. Nothing is uploaded, and nothing leaves the page until you press an export button.
2. Quick start
+- Pick a base colour and leave the grating at 4 cycles per degree.
- Set the viewing distance and pixel pitch to your own screen. Everything else on the page is derived from these two, so nothing is in degrees until they are right.
- Choose a map under Show — start with Retina, on-centre, then LGN, L − M.
- Read the scale bar under the map. Every map is normalised to its own extremes, so the numbers at the ends are what a bright pixel means.
3. The stimulus and the geometry
+A spatial frequency in cycles per degree is not a property of an image; it is a property of an image at a distance. The page asks for a viewing distance and a pixel pitch and derives every frequency from them — the grating and the V1 bank both, in the same unit.
| Control | What it does |
|---|---|
| Stimulus | A grating, or the built-in test card. The grating is the only source with a defined spatial frequency. |
| Chromatic / luminance | A chromatic grating holds L + M constant and swings the opponent axes; a luminance grating swings L + M. Try the retinal map under each: a luminance-driven centre-surround cell has little to answer to in an isoluminant pattern, and the page says so rather than showing an unexplained flat frame. |
| Viewing distance | Default 57 cm, where one degree subtends very nearly one centimetre — which is why it is the number vision labs use. |
| Pixel pitch | Default 0.248 mm, a typical desktop display. Your own is your screen’s width in millimetres divided by its width in pixels. |
| Spatial frequency | In cycles per degree. Above the pixel grid’s Nyquist limit the page tells you rather than drawing an alias. |
| Contrast, pupil | Contrast scales the grating. Pupil is kept in the saved state, but no stage reads it yet, so the sensitivity peak stays at 4 cycles per degree — in the shipped page it moved that curve’s peak from 9 to 104 cycles per degree. |
4. The five stages
+| Stage | What it computes, and what it rests on |
|---|---|
| Retina | A difference of Gaussians, giving on-centre and off-centre maps. The sigmas are in pixels, not degrees, and are this tool’s own. Below a surround weight of 1 the kernel is unbalanced and carries a constant term, which is why the off-centre map can be empty while the on-centre one is not. |
| LGN | Three opponent channels: L − M, S − (L+M)/2 and L + M. A neutral sits at the origin of both chromatic channels — that is the check they have to pass, and the shipped page failed it. |
| V1 | A Gabor bank: an even and odd quadrature pair per orientation, combined into a complex-cell energy map, plus a hypercolumn map of the preferred orientation at each pixel. Its frequency is in cycles per degree, the same unit as the stimulus. |
| V2 | The gradient of the V1 energy map, divisively normalised — a boundary signal. V2 does more than this, and the register says so. |
| V4 | A projection onto one of six directions in the opponent plane, half-wave rectified and normalised. It inherits the cone transform and the opponent axes above it. |
Each of the five switches really switches, and V2 needs V1. The shipped page had two more checkboxes — “enable ganglion / V1 proxies” and “enable the cortical Gabor bank” — which were read into state and consulted by nothing.
5. Reading a map
+- The scale bar is the map’s meaning. Every map is normalised to its own extremes, so without the numbers at the ends a bright pixel could be 0.2 or 20. The checkbox that hides it works; in the shipped page it was read into state and never consulted, and no map drew a bar to hide.
- A flat map is a result. When a stage finds no variation the bar says so rather than showing zero to zero.
- The hypercolumn map is a colour wheel: hue is the preferred orientation from 0° to 180°, and saturation is how selective that pixel is. Its scale bar shows the wheel.
- Colourmaps are viridis, inferno and grayscale. Grayscale is the honest one for a printout.
- Double-click either canvas to enlarge it.
6. The readouts
+- L − M peak gap — how far apart the two cone curves peak in the table this page holds. The published fundamentals peak 25 nm apart; the shipped table’s peak together, which is why it is marked a stand-in.
- Neutral off-axis — the largest departure a grey ramp makes from the origin of the opponent plane. It has to be essentially zero, and reads 2.2×10−4, which is the published matrix’s own rounding.
- Sensitivity and the curve’s peak — a bandpass weighting, this tool’s own, peaking at 4 cycles per degree. It is not a detection probability, and the page no longer prints one.
- Degrees per pixel and Nyquist limit — what your stated geometry works out to, and the highest frequency the pixel grid can carry.
- Frame RMS contrast — of the stimulus as drawn.
7. Installing a cone table
+The spectral curves this page draws are its own, and the Data rail takes a
real table: four columns — wavelength, L, M, S — as CSV or
JSON, with wavelengths increasing. Drop one in and the figure, the peak
readouts, the register entry and the reportable flag in every
export change by themselves.
A table whose L and M peak within 5 nm of each other is refused, because that is the defect being replaced. Nothing else on the page promotes a status: prose about a dataset never changes what it is.
8. Exports and sharing
+| Action | Result |
|---|---|
| A list of colours | Hex, one per line, through the front end the Lab is running: cone signals and all three opponent channels. Put a grey among them — it is the row that says whether the axes have their origin in the right place. |
| Download the CSV | cones-opponent.csv, under
a comment header naming the chain, the cone table and its peaks,
the opponent axes, your geometry, and whether anything in the file
may be reported against a publication. |
| Save the map | The active stage map as a PNG. |
| Download the settings | The same provenance as JSON. |
| Copy a link to this view | A URL carrying the stimulus, the geometry, every stage parameter and the map on screen. |
9. Library saves
+Library registers Visual Pathway as a Perception tool with preset capture.
The shared runtime hook reads
window.AANeuroscience.getState() and restores through
window.AANeuroscience.restoreState(saved).
10. What this tool does not hold
+- This is an educational simulator, not a physiological instrument. Nothing it prints is a measurement of any real visual system.
- The spectral table is this tool’s own. Its L and M curves peak at the same wavelength where the published ones peak 25 nm apart. It draws the cone figure and nothing else, and the Data rail takes a real one.
- The opponent axes are not DKL. Derrington, Krauskopf & Lennie’s space is built on cone contrast against a stated adapting white with isoluminant chromatic axes; these are raw differences of cone signals against L + M.
- The sensitivity curve is not a threshold model. No detection probability is reported from it.
- There is no spike raster. The panel that claimed one plotted image position on its horizontal axis and added uniform noise it called Poisson. Nothing here may be reported as a spike train or a firing rate.
- The retinal, V1, V2 and V4 constants are all this tool’s own: σ in pixels rather than degrees, and no stated retinal eccentricity.
The Data tab carries all of this entry by entry, with a status of verbatim, computed, stand-in or absent, and says what each one would take. For architecture and implementation details, see the Visual Pathway Developer Reference.