Animal vision: see the world through 69 animals’ eyes

Why a dog loses the red ball

Throw a red ball onto a green lawn and you can follow it all the way. Many dogs run straight past it and then find it with their nose. The ball has not become invisible. To a dog, red and green fall on the same kind of cone cell, so the ball and the grass come out as two shades of one dull olive. Throw a blue ball instead and the dog sees it at once.

That small scene holds most of what this exhibit is about. An eye is not a camera that records the world as it is. It samples light with a handful of receptor types, at a certain sharpness, at a certain speed, over a certain angle, and only above a certain brightness. Every animal has settled each of those differently. The stage redraws one garden, a night forest, a reef and a meadow seen from the air through the eyes of 69 animals, and every number it shows comes from a measurement in a paper or a university page listed at the end. Where nobody has measured something, the stage says "no figure" instead of guessing.

How an eye turns light into colour

The retina has two kinds of light sensors. Rods work in dim light and do not tell colours apart. Cones work in daylight and come in a few types, each tuned to a band of wavelengths. No single cone knows what colour it sees: it only reports how much light it caught. The brain compares the reports from different cone types, and that comparison is what we experience as colour.

People usually have three cone types. Measured one cell at a time in a human retina, their peaks sit near 420 nm (violet-blue), 534 nm (green) and 564 nm (yellow-green). The green and the long cone overlap a lot, but a red and a green object still give them different ratios, so we see two colours. Two lights that give all three cones the same catches look identical to us even when their spectra differ, and that is why a screen can fake almost any colour with only three narrow primaries.

The shape of a receptor's sensitivity curve is much the same from pigment to pigment once you know where it peaks. Govardovskii and colleagues fitted a template in 2000 that predicts the whole curve from the peak alone, including a small second bump at short wavelengths. The curves in the stage readout are drawn with that template, one per receptor type, and the colour model uses the same curves.

Two cones: dogs, cats, horses and the farm

Most mammals have two cone types: one for short wavelengths and one for middle-to-long ones. With two cones the eye can still tell blue from yellow, but red, orange, yellow and green all fall mostly on the long cone and differ mainly in brightness. A person with red-green colour blindness sees in a similar way.

In dogs, tests on trained animals published in 1989 found two cone pigments near 429 nm and 555 nm. Cats have the same pair, near 450 and 556 nm, measured in retinal recordings in 1969 and 1979. Horses have theirs at 428 and 539 nm, measured in 2001. Cattle, goats and sheep were tested together in 1998: a short cone between 444 and 455 nm and a long one between 552 and 555 nm. White-tailed deer have a short cone near 450 to 460 nm and a middle one near 537 nm.

Pick the dog and turn on "Half and half". The left half is your view, the right half the dog's. The red and green balls, more than 100 units apart for you on the CIELAB colour-difference scale, come out about 5 units apart for the dog. The blue ball stays more than 100 units away from the grass. If you buy a toy for a dog that plays on grass, blue or yellow is a better choice than red. Try the horse, the cow and the deer too: their long cones sit at slightly different wavelengths, so the reds turn olive by slightly different amounts.

Ultraviolet: mice, bees, budgerigars and goldfish

Many animals see a band of light we cannot: ultraviolet, below about 400 nm. Our own lens absorbs most of it before it reaches the retina. In 1991 Jacobs and colleagues found that some rodents have a cone most sensitive to ultraviolet, the first such receptor known in a mammal. Mice have one peaking at 360 nm and a green one at 508 nm, and most of their cones carry both pigments at once. Rats also have an ultraviolet and a middle-wave cone and, in tests published in 2001, used them together to tell colours apart.

Honeybees have ultraviolet, blue and green receptors, peaking at 344, 436 and 544 nm, and no receptor for red. Budgerigars and zebra finches have four cone types, one of them a true ultraviolet cone at 360 to 380 nm. Goldfish have four as well, at 356, 447, 537 and 623 nm. A flower bat from Central and South America, Pallas's long-tongued bat, responds to ultraviolet down to 310 nm, but in its case one pigment does both jobs: the green pigment's small second absorption band catches the ultraviolet, so the bat sees ultraviolet as brightness, not as a separate colour.

Ultraviolet patterns are common on flowers. Black-eyed Susans look plain yellow to us, but their petals absorb ultraviolet near the centre and reflect it at the tips, which makes a dark ring around the middle of the flower. In a 2014 field study, wild bees visited flower heads whose ultraviolet ring had been enlarged more often than ones whose ring had been made smaller. In the garden, pick the honeybee, the mouse or the budgerigar and look at the yellow flowers close up. Your screen cannot emit ultraviolet, so the stage carries a separate ultraviolet value for every surface and shows it in false colour; with "Ultraviolet only" you see that channel alone.

More receptors is not always better colour

The Japanese yellow swallowtail butterfly has eight varieties of receptor in six classes. In behavioural tests published in 2008, butterflies trained to one colour could tell it from another only 1 nm away at three places in the spectrum, near 430, 480 and 560 nm. That pattern is the mark of four receptor types working together, so the swallowtail sees with four channels even though it has more receptor types than that.

Mantis shrimp are the extreme case. Some species have 12 receptor types spread from 300 to 720 nm. One might expect them to tell colours apart better than anything else alive. Thoen and colleagues tested that in 2014 and found the opposite: the shrimp confused lights 12 to 25 nm apart, coarser than the butterfly and coarser than us. Their suggestion is that the shrimp do not compare receptors finely but recognise a colour by which narrow receptor responds, scanning the eye across an object. The stage models that idea with a simple rule: the strongest of 12 receptors names the colour, so the reef breaks into flat bands of hue. The same animals can also see circularly polarised light, which Chiou and colleagues showed in 2008.

One cone: sharks, octopuses and cuttlefish

Hart and colleagues measured the visual pigments of 17 shark species in 2011. Every one had a single cone type, peaking between 532 and 561 nm. With one cone an animal can judge brightness but not colour, so sharks may well be colour-blind. Whales, dolphins and seals also have a single cone type; they reached the same design on their own.

The common octopus has one visual pigment, peaking at 475 nm, and most behavioural tests find it colour-blind. Cuttlefish are similar, with one receptor type near 500 nm, yet they change the colour of their skin to match the background. Marshall and Messenger called this "colour-blind camouflage" in 1996. What cephalopods have instead is polarisation vision: their receptors sit at right angles to each other, and each absorbs light polarised in one direction best. Under water, where light reflected off silvery fish is strongly polarised, that is a useful signal. Pick the octopus at the reef and turn on "Polarisation": strongly polarised light shows yellow and unpolarised light blue.

How sharp: cycles per degree

Sharpness is measured with stripes. Show an animal pairs of black and white stripes and make them finer until it can no longer tell them from plain grey. The finest pattern it still sees, counted as stripe pairs (cycles) per degree of view, is its acuity. A person with 20/20 vision resolves about 30 cycles per degree; with the optics of the eye bypassed by interference fringes, people resolved gratings of about 60 cycles per degree in a 1965 study. The stage uses 60 as the person's limit and converts any figure into a Snellen fraction with 20/(600 ÷ cycles per degree).

Dogs reach 11.6 to 12.6 cycles per degree in recordings from brain and retina, about a fifth of the best human figure. Horses reached 23.3 at best in behavioural tests. Cats are given as 3.5 to 9, mice about 0.5 and rats about 1. A tawny owl tested in bright light beside two people saw less detail than they did and about as much as a pigeon. The wedge-tailed eagle is at the other end: 132 to 143 cycles per degree in Reymond's 1985 tests, more than twice the human figure and close to the 157 its pupil can let through.

Insects build their eyes from many small facets, each looking in its own direction. The angle between neighbouring facets sets how much detail they can pick up. It ranges from tens of degrees in primitive insects to 0.24 degrees in dragonflies, the finest known. The front of a honeybee's eye resolves about 0.54 cycles per degree. The principal eyes of the jumping spider Portia are tiny telephoto cameras with receptors 2.4 minutes of arc apart, about 12.5 cycles per degree, much better than any insect of its size. Scallops, with up to 200 small eyes along the edge of the shell, resolve about 2 degrees.

To show acuity on a screen, the stage blurs the picture by an amount that depends on how many pixels one degree of the view covers. A grating of c cycles per degree needs two pixels per cycle, so a view with p pixels per degree can show at most p ÷ 2 cycles per degree. In the garden one degree is only about 16 to 20 pixels tall on a 1440 px wide window and about 8 to 10 on a phone held upright, which is enough for 4 to 10 cycles per degree. That is coarser than a dog's eye, so at the normal view the dog's blur works out to a third of a pixel and the dog looks as sharp as you. The picture is not wrong: the screen itself is coarser than both eyes.

"Look far" fixes that the way a telescope would. It narrows the camera's view by 2, 4 or 8 times, so the same degree is spread over 2, 4 or 8 times as many pixels, and the blur is drawn at its true size for that view. At ×4 on a laptop one degree covers about 70 to 80 pixels: your half stays sharp, the dog's half blurs by about 2.5 to 3 pixels and the honeybee sees only a few facets. On a phone use ×8, where one degree covers about 70 pixels. The readout shows the pixels per degree and the blur at the current magnification. An eye sharper than yours, such as the eagle's, cannot be shown at all: the picture still has to pass through your own eyes, so the stage leaves it unblurred and says so.

The sky scene does the same with distance. It shows a meadow through a view only 4 m tall, like a telescope, at a distance you choose from 20 m to 1.5 km. At about 480 m a 40 cm rabbit spans about 3 cycles for you and about 7 for the eagle: a blob for you, still a rabbit for the eagle. Two rabbits sit 2.4 m apart so that with "Half and half" each side of the picture has one.

How fast: flicker fusion

Shine a light that switches on and off and raise the rate. At some frequency the flashes merge into a steady glow. That frequency is the critical flicker fusion frequency, and it says how quickly an eye can follow change. Inger and colleagues collected measurements for 81 species in 2014. People sit at about 60 Hz, cats at 55, dogs at 80, pigeons between 77 and 143 depending on the study, chickens between 72 and 105, and blowflies, honeybees and dragonflies at 200 to 300. Sharks measured so far sit between 18 and 37 Hz.

The lantern in the garden and the lamp post in the forest flicker at the rate you set, from 20 to 320 Hz. A screen cannot show a light flickering 100 times a second, so the lantern pulses slowly when the animal you picked could see its flicker, and the pulse grows deeper the further the lantern's rate is below the animal's fusion frequency. At 100 Hz, the flicker of a lamp on a 50 Hz mains supply, the lantern is steady for you and pulses for a bee. At 60 Hz, the refresh rate of many screens, it stays steady for a cat and pulses for a dog.

The numbers side by side

The atlas under the controls lists all 69 animals with their sources. Here are the ones with the most measurements, in the units the stage uses. A dash means no figure was found in these sources.

Read across a row and you can see the trade-offs. The eagle has the sharpest eyes but no special night ability on record; the cat sees in dim light but blurs fine detail; the bee sees ultraviolet and fast flicker but resolves only about half a cycle per degree.

Seeing at night

In dim light the cones stop working and the rods take over. Rods are more sensitive, but there is one kind, so colour fades away. People lose colour at about 0.02 cd/m², a level around twilight. Cats see in dimmer light than we do: tested with the same patterns, they were 0.74 log units, about 5.5 times, more sensitive, and most of that came from their larger pupils and the light-gathering of their optics. Barn owls have tubular eyes with an f-number of 1.13; a person's eye in the dark is about f/2.4, so the owl's image on the retina is about 4.5 times brighter. The stage uses that ratio as the barn owl's night gain and labels it as a model.

Two animals keep colour where we cannot. Helmet geckos have no rods at all, only three kinds of large cone, and in 2004 they told blue from grey at 0.002 cd/m², dim moonlight. Elephant hawkmoths chose coloured targets at 0.0001 cd/m², dim starlight, in tests published in 2002; in the same light people are completely colour-blind. Set the night forest to "Dim starlight" and try them. For these two the stage brightens the picture by the ratio of their colour limit to ours (10 and 200 times), and marks that as a model too.

Many night-active animals have a mirror behind the retina, the tapetum lucidum. Light that passes the receptors without being caught is sent back through them for a second chance. Dogs, cats, horses, cattle and many other mammals have one, and so do sharks. It is why their eyes shine back in a torch beam. Turn on the torch in the forest and look at the deer.

How wide: fields of view

Eyes on the front of the head see the same scene twice, which helps judge depth; eyes on the sides see more of the world at once. A person sees about 90 degrees to each side, about 180 degrees together, with about 120 seen by both eyes. A dog's field is about 240 degrees with 30 to 60 degrees of overlap, a cat's about 200. A horse covers about 350 degrees and is blind only straight behind and right in front of its forehead. Each eye of a tawny owl covers only 124 degrees, the narrowest field measured in any vertebrate, so its total field is 201 degrees with 48 degrees of overlap. Chameleons move their two eyes independently while they search, then lock both onto prey.

The panorama shows the whole circle around the camera, with your forward direction in the middle. The animal's field covers the middle of the strip; whatever lies outside it is hatched. For animals with no field figure the panorama shows all 360 degrees and says so.

Pupil shapes

Pupils come in round, slit, bar, W and pinhole shapes. Banks and colleagues compared 214 land animals in 2015 and found a pattern. Grazing prey animals such as goats, sheep, horses and deer have horizontal pupils, and their eyes roll as the head tips down to graze so the pupil stays parallel to the ground. That gives a sharp, wide view along the horizon. Small predators that ambush from close to the ground, such as cats, tend to have vertical slits, which help judge distance by blur. Cuttlefish have W-shaped pupils that even out light coming from above and below. The pupil of a helmet gecko closes to a row of pinholes.

The pupil sets the shape of the blur for anything out of focus. Turn on "Pupil blur" and pick the goat: objects in front of and behind the point the camera turns around smear sideways, while vertical edges stay sharper. The cat's slit does the opposite. The blur strength here is a model chosen so the shape shows clearly; the shape is the animal's.

Under water

Water absorbs light, and red goes first. Pope and Fry measured pure water in 1997: at 650 nm it absorbs 0.34 per metre, at 450 nm only 0.0092, about 37 times less. In the reef scene the light travels from the surface down to the reef and then across to your eye, and each colour is weakened along that path. At 15 m only about 2 per cent of the red is left, so the red fish turns grey-brown even for you, while blue and yellow fish keep their colour longer. Real sea water also scatters light and carries plankton; the stage uses pure water and a simple haze.

Heat: pit vipers

Pit vipers, pythons and boas have pits on the face that detect infrared from warm bodies. Bullock and Cowles recorded from the nerves of the pit in 1952 and showed that it responds to warmth nearby. In 2010 Gracheva and colleagues found the molecule behind it: the nerve fibres supplying the pits carry TRPA1 channels, which open when warmed. Infrared warms the pit membrane, and the channels report the change. It is a sense of heat, not a third kind of eye. The stage paints warm surfaces in false colour over a dim view: the tea glows, the sun-warmed stone glows faintly, and the open sky is cold.

The eye lab

The fifth scene puts the eyes themselves on a table. A human eye cut in half shows the cornea and lens bending light to a point on the retina, the image upside down, with the fovea where the cones are packed most densely. A compound eye is a dome of hexagonal facets; part of it is cut away to show the long ommatidia, each with its own lens and receptors. A third eye has a tapetum behind the retina, with a ray going in and coming back out. Five irises show the pupil shapes. The last model is a scallop eye, which focuses with a curved mirror instead of a lens. Palmer and colleagues showed in 2017 that the mirror is tiled with square guanine crystals and sends light onto two retinas, one for the centre of the view and one for the edge. Pick an animal while you are in the lab and the camera moves to its eye type.

Sixty-nine animals and the eyes they share

Version three of this exhibit more than doubles the cast, from 31 to 69 animals in ten groups. The new ones were chosen because people ask about them: farm animals and pets, zoo animals, birds of the garden and the shore, reptiles and frogs, the insects that bite or pollinate, and a few sea creatures whose eyes are odd enough to be worth a look. Each was added only with a paper that measured something about its eye, and the number on the stage links to that paper.

Some animals have not been measured directly but clearly share an eye system with one that has. The atlas marks them "grouped with" another animal, and the reason for grouping is itself a source. A chimpanzee is grouped with a person, because studies of primate photopigments find that Old World monkeys, apes and people all have three cone types with nearly the same peaks. A seal is grouped with the bottlenose dolphin: antibody staining found no blue cones in any of the five seal and sea lion species or the seven toothed whales examined, while their land relatives kept them. Pythons and boas are grouped with pit vipers, because the same heat-sensing channel was found in the pit nerves of all three. A giraffe is drawn with the cow's two cones, since most hoofed mammals tested have two, but its acuity of 25 to 27 cycles per degree is its own, counted from its retina. A grouped animal always uses the representative's curves unchanged, and the stage tells you so.

Breeds are a common question. The dog's two cone types come from studies of dogs in general, and no study in these sources compares colour vision between breeds. What does differ is where the dense cells sit. McGreevy and colleagues found that long-nosed dogs have a horizontal streak of ganglion cells like a wolf, while a pug has a single central spot. A greyhound and a pug probably see the same colours but scan the world differently. For cats the sources hold no breed comparison at all.

Elephants turned out to be a neat case. Pigments made from African and Asian elephant genes peak at 419 and 552 nm, the same pair as people with deuteranopia, one kind of red-green colour blindness. Pigs (439 and 556 nm), guinea pigs (429 and 529 nm) and rabbits (425 and 523 nm) are dichromats too, each with slightly different peaks. Old World monkeys sit near 430, 535 and 565 nm, almost exactly where our own cones are.

Birds bring four cones and coloured oil droplets. A European starling has single cones at 563, 504, 449 and about 362 nm, the last a true ultraviolet cone. A mallard's are at about 420, 452, 502 and 570 nm. A peafowl's pigments at 424, 458, 505 and 567 nm become peak sensitivities of 432, 477, 537 and 605 nm once the droplets filter them. The red-eared slider, a turtle, combines four pigments with droplets into seven cone types. Penguins went the other way and kept three cones, at 403, 450 and 543 nm, which suit blue-green water.

Seeing from an animal's height

The new Eye level switch moves the camera to where the animal's eyes would be and lets you look around from there. A mouse looks out from 3 cm, below the tips of the grass, so the lawn becomes a forest of blades and the apples on the plate stay hidden until you turn towards them. A dog looks from 50 cm, a person from 1.6 m, and a giraffe from 5 m, down on the whole garden. Pollinators start beside the low yellow flowers, a bee at 12 cm.

These heights are model values, typical heights for an adult standing or perched, not measurements from papers. Animals that live in water have no eye height on this stage, because the eye-level view works only in the garden and the forest. Height changes more than you might expect. At 3 cm everything near the ground is large and close, the sky fills the top half of the view, and an object a metre away can hide behind a single blade. At 5 m the same garden is a small patch whose flowers are dots, and the 25 to 27 cycles per degree of a giraffe's retina matter more than its colours.

Counting facets

A compound eye samples the world through many separate facets, each pointing in a slightly different direction. The angle between neighbouring facets limits how fine a pattern the eye can resolve, and the stage draws the view as a honeycomb with that spacing, each hexagon averaging its own patch. A worker honeybee has on average 5,432 facets in each eye, counted by Seidl and Kaiser in 1981, and the two eyes together see almost a complete sphere. Spread 5,432 hexagons over about half a sphere and their average spacing comes out near 2 degrees. At the front, where a bee looks at flowers, the facets point closer together, about 0.93 degrees apart. The readout gives that number and how many facets fit across the current view.

The mosquito shows how coarse an insect eye can be. Female Aedes aegypti facets are 6.2 degrees apart, so the smallest angle it resolves is about 12.3 degrees, roughly the width of your hand at arm's length. A later study of honeybee photoreceptors found that single receptors at the front of the eye respond to objects as small as 0.6 by 0.6 degrees, finer than the facet spacing suggests. The honeycomb is a lower bound on what a bee can detect, not a picture of everything it notices.

Not every compound eye works facet by facet. In some oceanic shrimps and lobsters, square tubes lined with mirrors reflect light from many facets onto one point and build a single bright image. Land described these reflecting superposition eyes in 1976. The stage cannot draw that difference, so the lobster gets the same honeycomb with a model spacing.

Grain in the dark

Light arrives as photons, and in dim light each receptor catches only a few of them in the time it takes to form an image. The count varies at random from moment to moment by about its own square root. A receptor that catches 100 photons varies by 10, which is 10% of the signal; one that catches 4 varies by 2, which is 50%. That randomness is photon noise, and it is why a moonlit scene looks grainy. The night views add grain that follows this rule: it grows with the square root of each spot's brightness and shrinks as the light level or the animal's gain rises. A cat, with a gain of 5.5 over a person, gets less grain in the same light. The scale of the grain is a model; the square-root rule is not.

Cold animals have an extra advantage. Rods occasionally fire with no light at all, set off by heat, and that dark noise sets a floor under the faintest light an eye can detect. Aho and colleagues compared species at different temperatures and found lower retinal noise and better absolute sensitivity in the cold, which helps toads and frogs. Frogs go further. They have two kinds of rods, and in a 2017 study frogs jumping toward light told blue from green right down to the faintest light they could see at all.

Eyes that barely make an image

The cast now includes animals whose eyes do something stranger than see a sharp picture. A nautilus has a pinhole eye with no lens, open to the sea, and its smallest separable angle is 5.5 to 11.25 degrees. A chiton has hundreds of tiny eyes in its shell plates with lenses made of aragonite, a mineral, which resolve about 9 to 12 degrees in air and in water. A box jellyfish has lenses that form an almost perfect image, yet the retina sits away from the focus, so each receptor sees a wide, blurred patch. A starfish has a compound eye without lenses at the tip of each arm, slow and colour-blind but good enough to steer back to its reef. At the other end of the scale, a giant squid eyeball about 27 cm across, with a 9 cm pupil, is the largest eye of any animal, and the rear eyes of a net-casting spider have an f-number of 0.58 and catch about 2,000 times more light per receptor than a jumping spider's.

A short history of measuring other eyes

For most of the 20th century, what an animal sees had to be worked out from behaviour: train it to choose one card for a reward and see which cards it confuses. Those experiments are slow, but they answer the question that matters, what the animal can actually use. Many of the numbers here still come from them: the horse acuity of 1992, the hawkmoth's starlight colours of 2002 and the mantis shrimp tests of 2014.

Physiology added a second route. In 1952 Bullock and Cowles put electrodes on the nerves of a pit viper's facial pit and watched them fire as a warm object came near. Microspectrophotometry lets researchers shine a tiny beam through a single receptor and read off its pigment's absorption, which is how the human cone peaks of 1980, the bird cones of 1997 and the shark pigments of 2011 were measured. Electroretinograms, recorded from the surface of the eye, gave the cone curves of horses, cattle and deer without dissecting anything.

Genetics is the newest route. Counting the opsin genes an animal carries shows how many pigments it could make: dragonflies turned out to carry 15 to 33. A gene count is not a colour sense, though. The mantis shrimp story is a reminder that behaviour still has the last word: many receptors do not by themselves give fine colour discrimination.

Trying it yourself

The panel under the controls has twelve tasks that ask you to use the stage the way a researcher would. Press Start on a task and it sets up a scene; then change the animal, the light or the distance until the box ticks. One asks you to find the flower pattern that only ultraviolet eyes see. Another asks for an animal that confuses the red and the green ball. Others ask who keeps colour by starlight, who can see behind itself, which animal sees a 60 Hz screen flicker, at what distance only the eagle still recognises a rabbit, how deep the red fish turns grey, what a horizontal pupil blurs, and how far you must look before a dog's view blurs next to yours. Three newer tasks use the eye-level view: find the nectar guide on the silverweed at a bee's height, find the apples as a mouse from 3 cm up, and visit three animals whose field reaches behind them. The case list below the tasks holds 102 studies; Show sets the stage to match each one.

How the views are computed

Everything happens in two passes on your graphics chip. First the scene is drawn into an off-screen image in linear colour, and a second, smaller image records for every surface its ultraviolet reflectance, its temperature, how strongly it polarises light, and how far away it is. Then one full-screen shader turns those two images into the animal's view.

Colour follows the approach of Machado and colleagues (2009), extended to other animals. The screen makes every colour from three narrow primaries, so each pixel is a known spectrum. Multiplying that spectrum by each of the animal's receptor curves gives the receptor catches, normalised so that white with average ultraviolet gives 1 in every receptor. Then the stage looks for the screen colour that gives you a similar result. For two receptors there is an exact answer: the colour on the plane between white and blue that gives the animal's two receptors the same catches, so any two colours the animal confuses come out identical. For three receptors that include ultraviolet, the shortest is shown as blue, the middle as green and the longest as red, the usual false colour for bees. Four-receptor eyes cannot be shown on a three-colour screen, so your colours are left as they are and ultraviolet, where the eye has it, is added as a purple tint. One receptor gives a grey picture.

Sharpness is a blur whose radius is the screen's pixels per degree divided by twice the animal's cycles per degree. Compound eyes are drawn as a honeycomb with the facet spacing, each cell averaging its patch. In dim light, brightness falls on a logarithmic scale with each animal's gain, colour fades below the animal's colour limit, and photon noise grows (see Grain in the dark). Under water each colour is multiplied by its transmission along the path. In the panorama the scene is rendered into a cube map and unwrapped around you.

How the garden is drawn

This version rebuilt the scenes so that their shapes look natural instead of cut from flat triangles. Each blade of grass is a narrow curved strip, thousands of them drawn in one batch, and a small program on the graphics chip bends them in the breeze so the processor does not move them one by one. Petals and leaves are curved surfaces rather than flat cut-outs. The silverweed flowers carry a pattern that feeds only the ultraviolet channel, so their dark centres appear for bees and other ultraviolet eyes but never in your own view. Stones are spheres bent by fractal noise with their surfaces kept smooth, and the apples, balls, plate and mug use finely divided shapes with a clear, glaze-like coat on top. Each scene lights itself from its own sky through a prefiltered environment map, shadows have soft edges, and on fast devices an extra pass darkens the creases where objects meet the ground (ambient occlusion). Slower devices drop the extra passes automatically and keep the same views.

These are pictures for comparing eyes, not photographs. The grass has no seeds, the deer has no fur, and the colours of real flowers vary far more than three beds can show. The ultraviolet patterns follow what the cited papers describe; they were not measured on these flowers.

What the model leaves out

A screen emits three colours chosen for human cones, so none of these views shows what an animal experiences; each one shows what its receptors would pick up, translated into colours you can see. The numbers marked "model" are choices rather than measurements: the receptor peaks for animals whose pigments were not in these sources, the brightness gains for the gecko and the hawkmoth, the pupil blur strength, the facet sizes of several insects and the eye heights in the eye-level view. Rods have their own spectral sensitivity near 500 nm, which shifts night scenes towards blue-green; the stage only fades colour. Brains matter as much as eyes: two animals with the same cones can still learn to read them differently. Individuals and breeds vary too, and the figures here are typical values from single studies.

Your own photo and camera

"My photo" and "Camera" load a picture from your device or open the camera, but only when you press them. The image is processed on this page by your graphics chip and is not uploaded or stored. A photo records only the three colours a camera sensor picks up, so it carries no ultraviolet, heat or polarisation; those channels stay empty and the stage says so.