A camera feed or a photo, shown with the colours a dog — or another animal — can tell apart.
- Running it — camera, photo, video,
--window,--species,--compare,--difference,--adaptation,--strength,--chroma-scale,--acuity,--fov. - The camera window — the species list, sliders, menus, keys, language and tooltips.
- Another GUI toolkit —
LiveSessionandrun(session). - Species — every preset, with its cone peaks and source, and how each sees a test chart.
- How it works — the model in five steps.
- Colour saturation —
fixedorrnl, and what the second rests on. - Acuity — blurring to what each species resolves, and when that shows.
- Map of differences — where two views differ noticeably, and by what measure.
- What it cannot show — limits, grouped by whether they can be lifted.
- Checking it —
--info,check_docs.py,ruff,lint-importsandpytest. - References
- Credits
- License
The dependencies are declared in pyproject.toml, and
uv is the only thing to install:
uv run sets up the rest on first use.
The window uses Tkinter.
The project asks uv for a Python already on the system, because the Tk in uv's own Python builds
lacks Xft and draws text without antialiasing; the window warns when that is the Tk it got.
A distribution's Python may need its Tk package (python3-tk on Debian and Ubuntu).
uv run dog-vision # live camera 0, as a dog
uv run dog-vision --camera 1 # another camera
uv run dog-vision --species horse # another animal
uv run dog-vision --species cow --compare horse # two animals side by side
uv run dog-vision --species cow --compare horse --difference # and where they differ
uv run dog-vision photo.jpg # writes photo.dog.png next to it
uv run dog-vision clip.mp4 # writes clip.dog.mp4 next to it
uv run dog-vision --window clip.mp4 # shows it in the window instead
uv run dog-vision --output-dir out photo.jpg # writes out/photo.dog.png
uv run dog-vision --adaptation 1 sunset.jpg./dog-vision in the checkout does the same from any directory, with the same
options; a link to it in a directory on PATH, such as ~/.local/bin, makes it a command.
--compareputs a second species where the original would be, with the same settings, so two animals can be told apart directly; with a photo, both go into the output side by side.--differenceadds a third image marking where the two differ noticeably; see Map of differences.--adaptation(0–1) adapts the cones to the scene instead of to daylight, so a sunset-lit scene loses its warm cast the way it would for an eye that has been in it for a while.--strength(0–1) blends the simulation with the original; 0 is the original.--chroma-scale fixed|rnlpicks how saturated a dichromat's colours come out; see Colour saturation.--acuityblurs to the species' visual acuity, and--fovsays how many degrees the image spans; see Acuity.- A video is converted frame by frame with the same options as a photo, at its own frame rate; see Video is written with the best method the system has.
--windowshows the photo or video in the window instead of converting it, and the window can convert it from there.--output-diris where a file converted on the command line goes, and where the window saves snapshots and recordings for this run; see Where files are saved.uv run dog-vision --helplists every option.
With ffmpeg installed, the sound of the original is carried over, and the video is encoded with the first of these that works on the machine:
- H.265 in software (
libx265) - H.265 in hardware: Apple VideoToolbox, NVIDIA, Intel Quick Sync, AMD
- H.264 in software (
libx264), then in the same hardware - MPEG-4 Part 2, which every ffmpeg has
- Each is tried on one frame first, since ffmpeg lists a hardware encoder whether or not its hardware is there.
- Without ffmpeg, OpenCV writes the video, with H.265, H.264 or MPEG-4, whichever its build has, and without sound: OpenCV does not handle audio.
- The colours are encoded and tagged as BT.709, the HD standard whose primaries sRGB shares, so that players do not read them as the older BT.601 and shift every colour.
- The command line and the window say which method was used, and whether the sound was kept.
The images are on the left, each with a line centred under it saying what it shows, and the controls
of the simulation on the right.
The window opens as large as the first image needs, and no larger than the screen.
The images follow the window's size; the controls keep theirs, as wide as with every section open,
so opening or closing one does not resize the images.
When the open sections do not fit the window's height, the controls get a scrollbar and the species
list keeps four rows.
View › Side panel or F9 hides the controls, and the images take their room.
What works on the window as a whole — the source, saving, the language — is in the menu bar, and a
status bar at the bottom says what is shown, how a conversion or a recording stands and what the
last action did.
The species, the View section's controls, the colour saturation and the acuity blur are in the
menu bar too, under Species and View, so they stay at hand with the side panel hidden.
The controls are grouped into sections, and a click on a section's title opens or closes it.
Acuity and View start closed, unless the command line turned on something in them
(--acuity, --compare, --difference).
| Control | Does |
|---|---|
| Species list or menu | picks the animal; the list takes arrow keys too |
| Selected species section | its cones, their sources and its RNL factors |
| mouse wheel over the list | scrolls it |
| mouse wheel over other controls | scrolls them, when they have a scrollbar |
| Adaptation to scene slider | same as --adaptation, in percent |
| Simulation strength slider | same as --strength, in percent |
| Colour saturation choice | same as --chroma-scale |
| Acuity box and slider | same as --acuity and --fov |
Side by side box, m |
a second image beside the simulation, or not |
| Left image choice | the original, or another species, as --compare |
Map of differences box, d |
a third image marking where left and right differ |
Reset button, r |
back to the values given on the command line |
File › Open file…, o |
a photo or a video instead of the camera |
| File › Camera | back to the camera |
| File › Convert file | writes the open file as shown, at full size |
File › Save snapshot, s |
saves dog-<species>-<time>.png |
File › Record video, v |
records dog-<species>-<time>.mp4; again stops it |
| File › Output folder… | where snapshots and recordings go, remembered |
| File › Converted files into … | puts converted files there too, remembered |
File › Quit, q, Esc |
quits |
| View › Side panel, F9 | hides or shows the controls on the right |
| Language menu | the window's language, starting as the system's |
Resting the pointer on a label, a box or a button shows what it means, in a few paragraphs. The facts under Selected species are text: a drag selects them, Ctrl+A selects them all, and Ctrl+C or a right click copies them.
- An open file is shown scaled down if it is large, so that the controls stay quick; Convert file works on it at full size, with the method described under Video is written with the best method the system has.
- A video plays at its own rate and starts over at its end.
- A snapshot is saved at the size the images are rendered, not at the size the window shows them: the camera's resolution, or a large file's scaled-down view.
- A recording is the view as shown, at that same size, and with every change made while it runs: another species, a slider, the left image. A video cannot change its size, so Side by side, Map of differences and the source are locked until it stops.
- A recording keeps the clock's pace: it is written at 30 frames per second, each frame repeated for as long as it was shown, whatever rate the camera delivers. It is encoded as a conversion is, on a thread of its own, and has no sound.
- Snapshots and recordings go to the output folder: the current directory until one is chosen
with File › Output folder….
A folder given with
--output-diris used for that run only; one chosen in the window replaces it and is remembered. - A converted file goes next to its original, as
<name>.dog.pngor<name>.dog.mp4. With File › Converted files into the output folder checked, it goes to the output folder instead, from the window and from the command line alike;--output-diron the command line always takes it. - A folder that does not exist yet is made when a file is saved in it.
- The window remembers both choices in
settings.json, in$XDG_CONFIG_HOME/dog-vision(~/.config/dog-visionwithout it) on Linux,~/Library/Application Support/dog-visionon macOS and%APPDATA%\dog-visionon Windows. A missing or damaged file means the defaults.
The window speaks English and Czech.
A language is one entry in LANGUAGES in i18n.py: its name, its
decimal point, and its translations, each keyed by the English text as in gettext.
A text reaches a translation through _(), or a helper such as text() for a widget's label; one
kept in a variable to be translated where it is shown is marked with N_(), as in gettext, so that
check_docs.py finds it.
The first of these that is set decides, and a language the window does not speak gives English:
LANGUAGE, which may list several languages, as incs:en_GB:enLC_ALLLC_MESSAGESLANG- on Windows, the user's interface language; elsewhere, the locale Python started in
So LANGUAGE=en uv run dog-vision opens it in English whatever the system says.
Only the window is translated: --help, --info and what a photo conversion prints stay English.
The window lives in gui/tk.py and does nothing but lay out widgets and
forward events.
Everything else sits in LiveSession in session.py:
render()returns the current view as an RGB array, orNonebefore the first frame. The camera or a video is read on a thread of its own, so a GUI can call it from any timer.open_file()andopen_camera()change the source;sourceis the open file, orNonefor the camera, andsource_name()says what is shown.convert_source()converts the open file in the background with the current settings;convertingandconversion_status()say how it stands.output_diris where snapshots and recordings go;set_output_dir()andset_convert_to_output_dir()change it and where converted files go, and save both tosettings, returning the error if they cannot.start_recording()andstop_recording()record the view asrender()shows it;recordingandrecording_status()say how it stands, and whilerecordingis true the window leavesside_by_side,differenceand the source alone.params(species, adaptation, strength, chroma scale, acuity, field of view),side_by_side,compare(the species on the left, orNonefor the original) anddifferenceare plain attributes to set.captions()says what each image of the rendered view shows, left to right.species_namesandchroma_scaleslist the choices in display order;species_labelsadds each species' kind of colour vision to its name.species_facts()returns what is known about the current species as (label, value, description) rows; a value is a tuple of the pieces of information it holds, such as a share and its source, so that a window breaks its lines only between them.languageis a code fromlanguages, which maps each to the language's own name, and sets the language ofspecies_labels,species_facts()andcaptions();translate()gives the window's own texts in it.reset(),save_snapshot()anderrorcover the buttons and a camera that stops.
A window in another toolkit is a module with the same run(session) function beside gui/tk.py;
main() in cli.py imports the window in one line.
Peaks are the wavelengths of maximum cone sensitivity as measured in each source.
A dichromat's longer cone is listed under L whatever it is called elsewhere, and a cone monochromat
has only that one, so it sees shades of grey.
The S-cone share is used only by the rnl colour saturation; where a
source reports a range across the retina, the middle of it is used, and a star marks a species
with no measurement found, which gets the assumed 10 %.
--species |
S [nm] | M [nm] | L [nm] | Peaks from | S cones [%] | Share from |
|---|---|---|---|---|---|---|
dog |
429 | – | 555 | Neitz, Geist & Jacobs 1989 | 10–18 | Mowat et al. 2008 |
cat |
450 | – | 550 | Guenther & Zrenner 1993 | 10–20 | Linberg et al. 2001 |
horse |
428 | – | 539 | Carroll et al. 2001 | 10–25 | Sandmann et al. 1996 |
cow |
451.3 | – | 555.3 | Jacobs, Deegan & Neitz 1998 | 5–10 | Schiviz et al. 2008 |
sheep |
445.3 | – | 552.2 | Jacobs, Deegan & Neitz 1998 | 5–10 | Schiviz et al. 2008 |
goat |
443.3 | – | 552.5 | Jacobs, Deegan & Neitz 1998 | 10 * | assumed |
pig |
440.7 | – | 556.7 | Jacobs, Deegan & Neitz 1998 | 5–10 | Schiviz et al. 2008 |
fallow-deer |
453.6 | – | 542.2 | Jacobs, Deegan & Neitz 1998 | 10 * | assumed |
white-tailed-deer |
456 | – | 536.8 | Jacobs, Deegan & Neitz 1998 | 10 * | assumed |
guinea-pig |
429 | – | 529 | Jacobs & Deegan 1994 | 10 * | assumed |
tree-squirrel |
444 | – | 543 | Blakeslee, Jacobs & Neitz 1988 | 10 * | assumed |
ground-squirrel |
436.7 | – | 518.9 | Jacobs, Neitz & Crognale 1985 | 7 | Kryger et al. 1998 |
ferret |
430 | – | 558 | Calderone & Jacobs 2003 | 7 | Calderone & Jacobs 2003 |
protanope |
420.7 | – | 530.3 | Stockman & Sharpe 2000 | 8–12 | Curcio et al. 1991 |
deuteranope |
420.7 | – | 558.9 | Stockman & Sharpe 2000 | 8–12 | Curcio et al. 1991 |
human |
420.7 | 530.3 | 558.9 | Stockman & Sharpe 2000 | 8–12 | Curcio et al. 1991 |
protanomalous |
420.7 | 530.3 | 548.9 | L shifted 10 nm, see text | 8–12 | Curcio et al. 1991 |
deuteranomalous |
420.7 | 540.3 | 558.9 | M shifted 10 nm, see text | 8–12 | Curcio et al. 1991 |
macaque |
431 | 536 | 565 | Bowmaker et al. 1978, 1991 | 10 * | assumed |
howler-monkey |
430 | 530 | 562 | Jacobs et al. 1996, S assumed | 10 * | assumed |
marmoset-female |
423 | 543 | 563 | Travis 1988, Williams 1992 | 10 * | assumed |
harbour-seal |
– | – | 510 | Crognale et al. 1998 | – | |
bottlenose-dolphin |
– | – | 524 | Fasick et al. 1998 | – |
Notes on the human and primate rows:
protanopeanddeuteranopeare humans lacking the L or the M cone; their values are the human pigment peaks.protanomalousanddeuteranomalousare anomalous trichromats of moderate severity: the L or the M cone is shifted 10 nm towards the other. Machado, Oliveira & Fernandes (2009) model severity the same way, with a 20 nm shift standing for dichromacy.humanis normal colour vision, and the reference thernlscale measures every other species against, so both its images are the original.marmoset-femaleis a female with the 543 nm and 563 nm variants of the M/L pigment; males, and females with two copies of one variant, are dichromats.- Every trichromat is given one L cone per M cone. Humans with normal colour vision range from about twice as many L as M cones to the reverse (Roorda & Williams 1999).
render_species_grid.py draws this image from the code, and
check_docs.py fails when it is out of date.
- They share the same two cone genes. Every dichromatic mammal here inherited one S and one L pigment from a common ancestor; species differ only in how those are tuned, by a few tens of nanometres at most.
- Cone sensitivities are broad, about 100 nm wide at half height, so a shift of 10 nm changes little.
- The visible differences are small but real: reds are darker for an L cone at shorter wavelengths (horse, ground squirrel), and the hue a species sees as white moves along the sweep.
- The large steps are between kinds of colour vision — trichromat, dichromat, monochromat — not between species of one kind.
The method is the one Brettel, Viénot & Mollon (1997) use for human dichromats:
-
Each cone is modelled with the Govardovskii et al. (2000) pigment template, from its peak alone.
-
The display is three Gaussian primaries, balanced so that white excites human cones like daylight.
-
That gives a matrix from linear RGB to the animal's cone excitations. For a dichromat it has one direction it cannot see: colours differing only along it look the same.
-
Each pixel is moved along that direction onto the plane R = G, which holds the grey axis and the blue–yellow axis. Neutral colours therefore stay neutral, and the animal's cones respond to the output exactly as to the input. A monochromat is moved onto the grey axis instead.
-
Adaptation and strength scale and blend the result.
-
The saturation of the result is set by one of two scales.
The whole transform is one 3×3 matrix on linear RGB.
The docstring of model.py states it as formulas.
Step 4 fixes which colours merge, but not how saturated the rest look: the plane has no natural
scale. --chroma-scale picks one of two:
| Scale | Saturation of the animal's colour axes |
|---|---|
fixed |
as the projection leaves it; the default, and what earlier versions always did |
rnl |
scaled so that a step the animal can just tell apart is one a human can |
A dichromat has one colour axis, blue–yellow. A trichromat has two, and for it fixed leaves the
image as it is, since nothing merges: rnl is the only way it looks different.
rnl counts just-noticeable differences (JNDs) the same way for the animal and for a human:
- Both are judged by the receptor noise limited (RNL) model of Vorobyev & Osorio (1998), around grey. It turns each cone's noise, set by the share of cones of that class, into a JND.
- The human is a normal trichromat looking at the output, with the
humanrow's cones and shares. - The output is rescaled until the two counts agree. The lightness is untouched.
A trichromat has two axes: blue–yellow keeps its hue, as in the
fixedscale, and changes only in saturation, while the other axis takes up the rest of the difference.--infoand the window show the factors for the species chosen. humantherefore comes out exactly as the original, and an animal shows how its discrimination compares with ours: below 1 it tells colours apart worse, above 1 better.
- RNL needs each cone's noise as a Weber fraction, and that is measured for almost no mammal. Taking the same value for animal and human makes it cancel, so none has to be chosen; the assumption is that it is the same.
- Some S-cone shares are assumed, marked with a star in the species table, and every trichromat's L:M ratio is.
- RNL is a threshold model. It is matched near grey; strongly saturated colours are extrapolated.
- Monochromats have no chromatic axis, so both scales give the same grey image.
--acuity removes the detail finer than the species resolves, the way AcuityView does
(Caves & Johnsen 2018): the image is filtered with the modulation transfer function
exp(−3.56 (f / acuity)²), f in cycles per degree, which is a Gaussian blur in linear light.
Acuity in cycles per degree, where a measurement was found:
--species |
Side by side [c/°] | One above another [c/°] | Source |
|---|---|---|---|
dog |
11.6 | 11.6 | Odom et al. 1983 |
cat |
10 | 10 | Wässle 1971 |
horse |
23.3 | 23.3 | Timney & Keil 1992 |
cow |
2.6 | 1.6 | Rehkämper et al. 2000 |
sheep |
12.8 | 12.8 | Sumita et al. 2013, 11.7 to 14 |
goat |
– | – | not found measured |
pig |
– | – | not found measured |
fallow-deer |
– | – | not found measured |
white-tailed-deer |
– | – | not found measured |
guinea-pig |
– | – | not found measured |
tree-squirrel |
2.8 | 2.8 | Jacobs, Birch & Blakeslee 1982, 1.8 to 3.8 |
ground-squirrel |
4 | 4 | Jacobs et al. 1980 |
ferret |
– | – | not found measured |
protanope |
72 | 72 | Land & Nilsson 2012 |
deuteranope |
72 | 72 | Land & Nilsson 2012 |
human |
72 | 72 | Land & Nilsson 2012 |
protanomalous |
72 | 72 | Land & Nilsson 2012 |
deuteranomalous |
72 | 72 | Land & Nilsson 2012 |
macaque |
60 | 60 | Nature Neuroscience 2024, about 60 |
howler-monkey |
– | – | not found measured |
marmoset-female |
30 | 30 | Troilo, Howland & Judge 1993 |
harbour-seal |
5.5 | 5.5 | Hanke & Dehnhardt 2009, in air |
bottlenose-dolphin |
3.66 | 3.66 | Herman et al. 1975, 8.2 arcmin stripes |
- Cattle resolve detail side by side better than one above another; their pupil is a horizontal oval, and the blur is stretched accordingly.
- A range in the source (sheep, tree squirrel) is replaced by its middle, and the source column says so.
- Species without a measurement are left sharp, and the window says so.
--fovis that angle, 60° by default, which is typical of a camera but only a guess for a photo. Halving it halves the blur.- An image shows the blur only if it has more pixels per degree than the species resolves. An image 640 pixels wide spanning 60° has about 11 pixels per degree, so a dog's blur is under half a pixel and invisible, while a 4000-pixel photo of the same scene blurs it by over two pixels.
- The result is right when it is seen at the angle it spans. Viewed smaller, the viewer's own acuity blurs it further; viewed larger, blur a human would not notice becomes visible.
--difference, or the window's Map of differences, adds a third image beside the two:
- Grey where the left and right images look the same.
- Red where they differ by more than one just-noticeable difference, deeper the larger the difference; the map's caption gives the share of pixels that do.
It compares the two images as a human sees them, in CIELAB, where one just-noticeable difference is about ΔE*ab 2.3 (Mahy et al. 1994). That catches differences in colour and in sharpness alike.
- A human looks at both images, so the map says what differs between two renderings, each already reduced to what its animal can tell apart.
- ΔE*ab 2.3 is an average. The true threshold varies across colour space, so the edge of the red region is approximate.
- Camera noise differs between a sharp and a blurred image, so with
--acuitynoise shows up as scattered red specks.
- Animals with an ultraviolet cone — mice, rats, birds, bees — are left out. An RGB camera records nothing of what that cone sees.
- For trichromats only the spacing of colours can change. The camera is itself trichromatic, so
two colours it records differently also excite a trichromat's cones differently: nothing merges.
The
rnlscale shows how far apart they look instead. - Colours a camera merges stay merged. Two surfaces that look alike to a human, and so to the camera, can differ for the animal.
- What the colours feel like to the animal is unknowable. The simulation shows which colours it cannot tell apart, and nothing more.
- The lens and the macular pigment are ignored. This is the likely reason the model puts the human deuteranope's neutral point well short of the measured value of about 505 nm, while it matches the dog's 480 nm (both from Neitz, Geist & Jacobs 1989).
- The display primaries are generic, not those of the screen in front of you.
- Blue–yellow is a convention. Which human hues stand for a dichromat's single chromatic axis is not fixed by physics.
- Colour and acuity are simulated — not motion sensitivity or dim-light vision.
uv run dog-vision --info --species <name>prints the derived matrices and checks that grey is preserved, that the output excites the cones exactly as the input does, and the neutral point.uv run tools/check_docs.pychecks this README against the code: the species table with its S-cone shares and sources, the relative links and anchors, the neutral points quoted above, and that the test chart and the apple figures are what the code renders (render_species_grid.py,render_photo_figures.py). It also checks that each translation ini18n.pynames every species, describes every label and translates every text the code hands over to be translated, and that every English text it translates still occurs in the code.uv run ruff checklints the code, and keeps its lines within 120 characters.uv run lint-importschecks that the code keeps its layers:dog_vision.coreimports neither the window nor the command line, anddog_vision.guidoes not import the command line.uv run pytestruns the tests intests/: the model's invariants (grey stays grey, with the fixed saturation the animal's cones see the output as they see the input, and the RNL one lets a human count the animal's differences), the image pipeline, the facts and their translations, converting and recording video, the command line,LiveSession, and the window, driven through its keys and menus. The window's tests run on a virtual display when Xvfb is installed and on$DISPLAYotherwise, and are skipped without either; the tests of sound, colours and padding in a video need ffmpeg.uv run pytest --covalso lists the lines and branches ofdog_visionthat no test reaches.
- Blakeslee, B., Jacobs, G. H. & Neitz, J. (1988). Spectral mechanisms in the tree squirrel retina. J. Comp. Physiol. A 162, 773–780.
- Bowmaker, J. K. et al. (1978). The visual pigments of rods and cones in the rhesus monkey, Macaca mulatta. J. Physiol.
- Bowmaker, J. K., Astell, S., Hunt, D. M. & Mollon, J. D. (1991). Photosensitive and photostable pigments in the retinae of Old World monkeys. J. Exp. Biol.
- Brettel, H., Viénot, F. & Mollon, J. D. (1997). Computerized simulation of color appearance for dichromats. J. Opt. Soc. Am. A 14, 2647–2655.
- Calderone, J. B. & Jacobs, G. H. (2003). Spectral properties and retinal distribution of ferret cones. Visual Neuroscience 20.
- Carroll, J., Murphy, C. J., Neitz, M. et al. (2001). Photopigment basis for dichromatic color vision in the horse. Journal of Vision 1.
- Caves, E. M. & Johnsen, S. (2018). AcuityView: an R package for portraying the effects of visual acuity on scenes observed by an animal. Methods Ecol. Evol. 9, 793–797.
- Crognale, M. A., Levenson, D. H., Ponganis, P. J., Deegan, J. F. & Jacobs, G. H. (1998). Cone spectral sensitivity in the harbor seal (Phoca vitulina) and implications for color vision. Can. J. Zool. 76, 2114–2118.
- Curcio, C. A. et al. (1991). Distribution and morphology of human cone photoreceptors stained with anti-blue opsin. J. Comp. Neurol. 312.
- Fasick, J. I., Cronin, T. W. et al. (1998). The visual pigments of the bottlenose dolphin (Tursiops truncatus). Visual Neuroscience 15.
- Govardovskii, V. I., Fyhrquist, N., Reuter, T., Kuzmin, D. G. & Donner, K. (2000). In search of the visual pigment template. Visual Neuroscience 17, 509–528.
- Guenther, E. & Zrenner, E. (1993). The spectral sensitivity of dark- and light-adapted cat retinal ganglion cells. J. Neurosci. 13, 1543–1550.
- Hanke, F. D. & Dehnhardt, G. (2009). Aerial visual acuity in harbor seals (Phoca vitulina) as a function of luminance. J. Comp. Physiol. A 195.
- Herman, L. M., Peacock, M. F., Yunker, M. P. & Madsen, C. J. (1975). Bottle-nosed dolphin: double-slit pupil yields equivalent aerial and underwater diurnal acuity. Science 189.
- Jacobs, G. H. & Deegan, J. F. (1994). Spectral sensitivity, photopigments, and color vision in the guinea pig (Cavia porcellus). Behav. Neurosci. 108, 993–1004.
- Jacobs, G. H., Birch, D. G. & Blakeslee, B. (1982). Visual acuity and spatial contrast sensitivity in tree squirrels. Behavioural Processes 7.
- Jacobs, G. H., Blakeslee, B., McCourt, M. E. & Tootell, R. B. H. (1980). Visual sensitivity of ground squirrels to spatial and temporal luminance variations. J. Comp. Physiol. A 136.
- Jacobs, G. H., Deegan, J. F. & Neitz, J. (1998). Photopigment basis for dichromatic color vision in cows, goats, and sheep. Visual Neuroscience 15, 581–584.
- Jacobs, G. H., Neitz, J. & Crognale, M. (1985). Spectral sensitivity of ground squirrel cones measured with ERG flicker photometry. J. Comp. Physiol. A 156, 503–509.
- Jacobs, G. H., Neitz, M., Deegan, J. F. & Neitz, J. (1996). Trichromatic colour vision in New World monkeys. Nature 382.
- Kryger, Z. et al. (1998). The topography of rod and cone photoreceptors in the retina of the ground squirrel. Visual Neuroscience 15.
- Land, M. F. & Nilsson, D.-E. (2012). Animal Eyes, 2nd edition. Oxford University Press.
- Linberg, K. A., Lewis, G. P. et al. (2001). Distribution of S- and M-cones in normal and experimentally detached cat retina. J. Comp. Neurol. 430.
- Machado, G. M., Oliveira, M. M. & Fernandes, L. A. F. (2009). A physiologically-based model for simulation of color vision deficiency. IEEE Trans. Vis. Comput. Graph. 15.
- Mahy, M., Van Eycken, L. & Oosterlinck, A. (1994). Evaluation of uniform color spaces developed after the adoption of CIELAB and CIELUV. Color Research & Application 19, 105–121.
- Mowat, F. M. et al. (2008). Topographical characterization of cone photoreceptors and the area centralis of the canine retina. Molecular Vision 14.
- Multiple loci for foveolar vision in macaque monkey visual cortex (2024). Nature Neuroscience.
- Neitz, J., Geist, T. & Jacobs, G. H. (1989). Color vision in the dog. Visual Neuroscience 3, 119–125.
- Odom, J. V., et al. (1983). Canine visual acuity: retinal and cortical field potentials evoked by pattern stimulation. Am. J. Physiol. 245.
- Rehkämper, G., Perrey, A., Werner, C. W., Opfermann-Rüngeler, C. & Görlach, A. (2000). Visual perception and stimulus orientation in cattle. Vision Research 40, 2489–2497.
- Roorda, A. & Williams, D. R. (1999). The arrangement of the three cone classes in the living human eye. Nature 397, 520–522.
- Sandmann, D., Boycott, B. B. & Peichl, L. (1996). Blue-cone horizontal cells in the retinae of horses and other Equidae. J. Neurosci. 16.
- Schiviz, A. N., Ruf, T., Kuebber-Heiss, A., Schubert, C. & Ahnelt, P. K. (2008). Retinal cone topography of artiodactyl mammals: influence of body height and habitat. J. Comp. Neurol. 507, 1336–1350.
- Stockman, A. & Sharpe, L. T. (2000). The spectral sensitivities of the middle- and long-wavelength-sensitive cones derived from measurements in observers of known genotype. Vision Research 40, 1711–1737.
- Sumita, S., Prescott, N. B., Broom, D. M., Wathes, C. M. & Phillips, C. J. C. (2013). Visual discrimination learning and spatial acuity in sheep. Appl. Anim. Behav. Sci. 147.
- Timney, B. & Keil, K. (1992). Visual acuity in the horse. Vision Research 32.
- Travis, D. S., Bowmaker, J. K. & Mollon, J. D. (1988). Polymorphism of visual pigments in a callitrichid monkey. Vision Research 28, 481–490.
- Troilo, D., Howland, H. C. & Judge, S. J. (1993). Visual optics and retinal cone topography in the common marmoset (Callithrix jacchus). Vision Research 33.
- Viénot, F., Brettel, H. & Mollon, J. D. (1999). Digital video colourmaps for checking the legibility of displays by dichromats. Color Research & Application 24, 243–252.
- Vorobyev, M. & Osorio, D. (1998). Receptor noise as a determinant of colour thresholds. Proc. R. Soc. B 265, 351–358.
- Williams, A. J. et al. (1992). The polymorphic photopigments of the marmoset: spectral tuning and genetic basis. EMBO J. 11.
- Wässle, H. (1971), as listed in the micaToolbox acuity list.
The apple photo is Shiny red apples by Leon Brooks, released into the public domain.
dog-vision is free software under the GNU General Public License, version 3 or any later version
(GPL-3.0-or-later); the full text is in LICENSE.




