A ritual has long taken root in my feed: whenever it serves up something about perception modeling, I can’t just scroll past. This time Habr brought me a case with a very old pedigree: a person going by ymik, who was writing on Habr about modeling human vision and color blindness seventeen years ago, returned to the subject and built a web model that shows ordinary phone video as it might look to a cat. Not a Photoshop preset called “mega cat filter,” but an attempt at an honest recalculation: dichromatic color, elliptical stretching of the field, blurred vertical periphery, and a special relationship with horizontal motion. The code and demo are open—you upload any clip and watch the world stop being human.
I went digging through the primary sources to see how much faith one could put in such a reconstruction. And here’s what’s interesting: almost every choice in this model rests on real research, some of it older than the researchers themselves. But the tastiest part is what the model does not—and cannot—contain.
Let’s start with color, because this is where the cat’s setup is rigidly defined. Humans have three types of cones; cats have two: with sensitivity peaks around 450–460 nanometers (the blue range) and 550–560 nanometers (the yellow-green range). Red and green do not exist as separate colors for the feline retina—they collapse into a single yellow-green channel. That makes the cat the closest relative of a person with protanopia, with one difference: for a cat, this “loss” is the species norm, not a defect.
How did anyone discover that cats are dichromats in the first place? The history of that knowledge is a pleasure in itself. In 1964, Science published a paper with the telling title “Color Discrimination in the Cat”: cats were trained to distinguish colored targets, and they managed it. But skeptics immediately asked: what does color have to do with it? A cat might distinguish “red” from “blue” not as colors, but as different brightnesses—after all, every colored surface reflects a different amount of light. And this is where the beautiful part begins: to prove that cats see wavelength itself rather than brightness, researchers in the Journal of Physiology (the paper came out in the late ’70s) trained cats to distinguish red from cyan, then started tormenting the stimuli: they showed them against a background bright enough to blind the rods—the receptors responsible for night vision. If the cat were distinguishing colors through rods, it would have fallen apart against such a background. It didn’t. One cat was additionally blasted for five minutes with light at 9000 candela per square meter—it continued to distinguish red from cyan. The conclusion: discrimination runs through cones, and there is more than one kind of cone. Neurophysiological confirmation followed: in the lateral geniculate body—the relay station between eye and cortex—researchers found rare color-opponent neurons receiving input from green- and blue-sensitive cones.
The practical consequence: a cat’s world is not gray. Blue objects remain blue, yellow ones remain yellow, but everything red-green is pulled into a yellowish-gray palette. That is why models recalculate the image as though the observer had different photoreceptors—the author honestly admits that his simplified model does not measure up to the Govardovskii spectral curves (the very ones that describe visual opsin pigments), but the principle is the same as in protanopia simulations: not “paint the image yellow,” but recalculate the channels’ contributions.
The second layer of difference is geometry. Human vision rests on the fovea: a tiny depression in the center of the retina, where cone density is greatest and where we truly “see.” The rest of the field is deep periphery, almost statistics. Cats have no fovea. Its role is played by the area centralis—a zone of elevated ganglion-cell density without a depression—plus an elongated horizontal band of increased density: the visual streak.
A band instead of a point is no accident; it is the evolutionary signature of a way of life. A 2025 review by Mass and Supin puts it plainly: in mammals, the zone of best vision comes either as a spot (predators), a horizontal band (animals of open spaces, for which the horizon is the main organizer of the scene), or a combination. In cats, it is a spot slanting toward a band: resolution is highest not in the geometric center of the field, but along the horizontal meridian, the very line where the entire terrestrial scene lives. The top and bottom of the frame are, for a cat, like the corners of our eyes: there is light, but no detail. Ymik’s model reproduces this with blurred vertical periphery—and it is not a decorative effect, but a direct projection of anatomy.
Now for the numbers, because this is where science speaks in competing voices. A cat’s field of view is around 200 degrees against our 180; some methods (adding together monocular fields) produce figures up to 270. The binocular zone—the region seen by both eyes at once—is narrower in cats than in humans: their eyes are set farther apart and not quite in the same plane, so stereopsis comes at the price of reduced overlap. Visual acuity, though, is more interesting: classic studies from the 1970s (Blake, Jacobson) consistently gave 20/100–20/200—that is, a cat sees at twenty feet what a human sees at one hundred to two hundred. That is “roughly 5–10 times worse than humans,” and veterinarians still cite those figures. But in 2013, Clark and Clark checked everything again: the same domestic cats, the same grating task, but without the old handicaps—unlimited time, high luminance, high contrast, room for normal accommodation. The result was 20/30, almost human-level (the norm is 20/20). Paul Miller, professor of veterinary ophthalmology in Wisconsin, commenting on this spread, said 20/30 seemed more plausible to him: “judging by the way cats stare intently at small insects on the ceiling.” The gap between the estimates is three- to sixfold, and the reason, it seems, is not that “science got it wrong”: for half a century, cats were tested within laboratory constraints that impaired their vision—cramped conditions, haste, dim targets. As soon as the conditions let them work at their own pace, acuity leapt toward the human benchmark. A fine example of how the measuring instrument itself becomes a participant in the experiment.
By the way, a cat’s near zone is weak: at about 25 centimeters and closer, the image goes blurry—a cat is literally farsighted up close, like a human around forty-five years old. The whiskers on its muzzle and front paws take the place of a close-up magnifying glass.
Now for the model’s most contentious claim: “the cat cortex especially emphasizes horizontal movement.” Habr’s phrasing sounds like a tall tale, so I went to check the primary sources—and found something more interesting than a simple “yes, it sees better.”
The cat retina has a systematic radial bias: outside the central zone, most ganglion cells respond best to stimuli oriented along the line from “receptive field to center of gaze” (Levick and Thibos, 1982). This bias is most strongly expressed along the horizontal meridian—that is, where the visual streak sits. The same principle is inherited by the striate cortex (Leventhal, 1983) and the extrastriate area 19 (Leventhal, Schall, and Wallace, 1984; 226 neurons): the relationship between “where the line is pointing” and “where its receptive field lies” persists throughout the visual tract and is strongest precisely on the horizontal meridian.
Let me translate that from neurophysiology. The entire visual system of the cat is tuned to a scene spread horizontally: ground beneath the paws, the horizon in the distance, prey running along the ground from left to right or right to left. Not “a cat sees all horizontal movement better”—rather, the processing of detail is concentrated where the world most often moves: along the ground. Vertical movement (a falling object, a droplet flying downward) is statistically less interesting to this system. In the model, this becomes its most ingenious trick: motion detection works relative to the background—first, camera shift is estimated and subtracted out (the wobble of the hands), then the remainder, the actual motion of objects, is used to make the decision. A fast horizontal object receives two privileges at once: time around it stretches (playback slows to 0.85×, while the ordinary scene runs 1.15× faster), and only it stays sharp while the background and periphery blur. The motion mask does not vanish instantly—it decays along a logistic curve over 150 milliseconds, like the continuation of a trajectory.
This horizontal priority has an optical accomplice as well. In 2015, Science Advances published a major paper by Banks and co-authors on why animals have different pupil shapes at all (214 species, multinomial regression). The vertical slit—the calling card of domestic cats—turned out not to be a random ornament: it creates astigmatic depth of field. Vertical contours at different distances remain sharp (the cat measures their distance through stereopsis), while horizontal contours outside the focus blur—and the degree of blur itself encodes distance. For an ambush predator that leaps at prey and has to assess, in a fraction of a second, “how far away it is,” this is the ideal optical configuration. The vertical slit clicks along the vertical axis of the world; horizontal defocus measures the horizontal—again, the same axis.
A cat’s night vision is usually presented as magic (“cats can see in the dark!”), but its mechanics are thoroughly engineered—and that is partly why they are beautiful. Three components. First: a cat has more rods in its retina than a human, and fewer cones—at night, sensitivity matters more than color. Second: the lens sits closer to the retina, and the cornea is larger—more light enters the eye. Third, and most famous: the tapetum lucidum, a reflective membrane behind the retina. A photon that misses the photoreceptors bounces off the tapetum and flies back—the retina gets a second chance to catch it. Hence the glowing eyes in flash photographs: the eyes are not glowing; the reflector is returning what the retina did not absorb.
The most unexpected thing about the tapetum is its chemistry. In most animals, the reflector is made of guanine crystals or zinc complexes, but in cats, the reflective component is riboflavin, vitamin B2. Riboflavin does not merely reflect: it is fluorescent, amplifying light at a particular wavelength, in the yellow-green range—precisely where the cat’s second (yellow-green) cone sits. A flashlight-reflector with chemical selectivity, built in behind the retina. By estimates (veterinary ophthalmologist Miller; consistent with dvm360’s figure of 16% illumination), a cat needs light 5.5–7 times weaker than a human does—not that it “sees in the dark,” but that it sees in the twilight that is already night to us.
The price for all this is well known: resolution. A small ganglion cell delivers high acuity, but a large, light-sensitive one does not; in serving the night, the cat pays in detail.
And here is the most honest part of the whole story. Ymik’s model is solid: dichromacy, elliptical compression of the field to fit our rectangular screen, vertical blur, motion with camera compensation, all in client-side WebGL shaders, without a single neural network (the author tried TensorFlow.js for optical flow—large models are too much for a browser, small ones are noisy; classical global-shift estimation proved more stable). The author announces the next steps: dog, bull, and then—tetrachromatic vision in birds and octopuses.
But this whole beautiful construction has a fundamental limit, and it is not computational. In 2014, Proceedings of the Royal Society B published a paper by Douglas and Jeffery: a cat’s lens transmits a significant part of ultraviolet light (UVA, 315–400 nanometers). The human lens is yellow and acts as a UV filter—we physically cannot see ultraviolet; it is blocked before it reaches the retina. A cat’s lens is transparent to UVA, giving cats probable sensitivity to a spectrum for which we have no words at all. Bird feathers carry UV patterns invisible to us; urine markings glow in UV; snow shadows are full of ultraviolet gradients. For a cat, the world probably simply contains more channels than ours.
You cannot translate ultraviolet vision into video not because the engineer lacked the skill, but because the viewer has no receptor into which that signal can be placed. Any simulation of another creature’s vision is a projection into our color cube, and that cube has three axes. Everything that lives outside the cube remains beyond the edge of the frame: the model can honestly show how a cat loses red-green discrimination, but it cannot show what the cat gains in return.
I like this asymmetry more than the model itself. The WebGL filter is the tip of work that has been going on for a long time: from engineering simulations of color blindness seventeen years ago to spectrally precise recalculations. And it honestly does the crucial thing—it makes you feel that “seeing” is not such an objective matter. Vision is not a photograph of the world, but an interface honed for a way of life: a band instead of a fovea—for the horizon; a slit instead of a circle—for ambush; riboflavin behind the retina—for twilight. Even if one channel remains forever beyond the frame.
Sources: