There was a similar web-game back in the day but for tones. You had to identify which one was higher or lower. Also gave a statistical comparison at the end. Near the end the difference was a couple hertz or something, so next to impossible as well.
I can't find it now. Is anyone else aware of what I'm talking about?
This is fun! I just played once and got 0.0016, which it says is "absurdly below the theoretical limit"...
Okay, tried again and got 0.0034 which is still says is beyond the human limit! I'll have to give this to my mum because we often argue about colours and I suspect she might be a tetrachromat.
Nice job. Kind of reminds me of this one which increases the number of squares with the odd-one out becoming more subtle as you progress further in the game, but I prefer your sliding mechanic better for this kind of game.
Ooh this one is fun too! Though it doesn't get quite as hard as the slider one. Breezed through all 47 levels of this pretty easily while there were one or two impossible seeming ones in the slider.
This is probably a fun way of finding the median screen quality (in a world where no one has color blindness or any eye issues, so scrape 15-18% of the results away ig)
The code contains a function that, given the target ΔE, generates two colors in floating-point Oklab representation, separated by that distance. But there is no check whether the two generated colors end up rounding to exactly the same one on 8-bit displays. So, I was asked to find a boundary (while the claim was that there were two distinct colors 0.0013 ΔE apart) between RGB(80, 83, 152) and RGB(80, 83, 152). Obviously unfair.
I hope you post this again when you do - I was presented with the "0.00080" difference a couple times, and it looks like this is where it becomes actually impossible because of this issue.
Are you using Oklab channels to measure delta-E / difference? If so, Oklab is a hacky way to approximate a real colorspace with just one matrix multiplication, the channels have no meaning and are not related to delta-E. Use real Lab*, it'll take 10 minutes with an LLM.
EDIT: Just read the blog post. I thought HSL was bad for design, Oklab is much worse. It just goes right ahead and reuses color science terms so it sounds it got it all right. (dEOK existing and its "JND" being 0.02 absolutely made my head spin. None of this is recognizable to a color scientist)
not knowing anything about color, i will admit i am a bit confused. i scored 0.0034 and was told "if you're not already calibrating displays for a living, you're leaving money on the table". which, to me, implied i did quite well!
but, reading the scores posted here, most people are doing a lot better than me. i doubt all of us are crazy good...
so, i assume the front page is a typo: "most people land around 0.02" (should be 0.002, not 0.02)? if yes, then i am back to not understanding the message i got about calibrating displays, because i did quite a bit worse than 0.002.
edit: nerd-sniping myself a little bit. but it appears (stressing: i know nothing) the "0.02" is accurate, but calculated by showing someone two colors and asking "are these different" until the person answers the question correctly 50% of the time. which is a different question than "where, precisely, is the line between these two colors". with the different question, it ends up compressing the result down by about an order of magnitude.
Right. The average score is under different test conditions. Obviously this game is a little silly version with very little accuracy to the lab testing, but hopefully it gets people thinking about this stuff a bit more! Which given your investigations into this, I would say it has succeeded.
I'm colorblind, but I ended up getting a 0.0028 "much better than average" score. Hmm... Fun site!
To promote some further reading:
OKLab isn't actually a perceptually uniform colorspace. It's better than others, but it was specifically chosen as a tradeoff between accuracy and speed (hence the name OK). When you start digging this deep, you quickly learn that we have yet to invent any perceptually uniform colorspaces; even the most precise models we have end up using fits and approximations. Color has some really inconvenient properties like depending strongly on brightness and background. Frankly, given the differences in human biology (having orders of magnitude differences in relative numbers of each cone, for instance), it's surprising we agree as much as we do! Human color perception is an endless pit of complexity.
(Note, I don't say any of this to detract from what you've built here, merely expand. Your site is awesome and I love it!)
If 0.02 is the JND of deltaEOK, how come everybody is getting results an order of magnitude smaller? Even the author himself (at https://www.keithcirkel.co.uk/too-much-color/) says they get 0.0028, but never elaborate on the significance of that result.
JND is an average. A lot of people will do a lot worse. The measure, as I understand it, is also under different test conditions, while this is a game where people are on their own and able to - for example - tilt their head trying to find the exact angle to see the difference.
This takes something as nerdy as decimal places in CSS colors and turns it into a fun, practical read. It feels like you’re being walked through the rabbit hole by a friend who’s done way too much homework, then hands you a few simple rules you can actually remember and use.
I can't find it now. Is anyone else aware of what I'm talking about?
But now I'm aware of a slightly darker vertical band on my screen, compared to the rest of the display. x_x
Okay, tried again and got 0.0034 which is still says is beyond the human limit! I'll have to give this to my mum because we often argue about colours and I suspect she might be a tetrachromat.
Both tests on a Pixel 10 btw
https://hnarcade.com/games/games/what-s-my-jnd
https://vectorization.eu/color-perception-test
But I think this kind of test can really be limited on your display and color profiles.
Most of my misses were on blues, but sometimes the line was obvious. Other times, I could “see” many phantom bars of slightly different colors.
And in bright sunlight, I can see variations in the film in my phone screen. I suspect this could affect a test like this.
Here's the related article on how much accuracy is really needed in CSS values. https://www.keithcirkel.co.uk/too-much-color/
The code contains a function that, given the target ΔE, generates two colors in floating-point Oklab representation, separated by that distance. But there is no check whether the two generated colors end up rounding to exactly the same one on 8-bit displays. So, I was asked to find a boundary (while the claim was that there were two distinct colors 0.0013 ΔE apart) between RGB(80, 83, 152) and RGB(80, 83, 152). Obviously unfair.
EDIT: Just read the blog post. I thought HSL was bad for design, Oklab is much worse. It just goes right ahead and reuses color science terms so it sounds it got it all right. (dEOK existing and its "JND" being 0.02 absolutely made my head spin. None of this is recognizable to a color scientist)
I tried it on a recent Pixel with brightness set to two-thirds, and this is my result:
https://www.keithcirkel.co.uk/whats-my-jnd/?r=ArggKP__c4_b
It helps if I move side to side like a party parrot. I'd love to see a histogram of where I stand.
not knowing anything about color, i will admit i am a bit confused. i scored 0.0034 and was told "if you're not already calibrating displays for a living, you're leaving money on the table". which, to me, implied i did quite well!
but, reading the scores posted here, most people are doing a lot better than me. i doubt all of us are crazy good...
so, i assume the front page is a typo: "most people land around 0.02" (should be 0.002, not 0.02)? if yes, then i am back to not understanding the message i got about calibrating displays, because i did quite a bit worse than 0.002.
edit: nerd-sniping myself a little bit. but it appears (stressing: i know nothing) the "0.02" is accurate, but calculated by showing someone two colors and asking "are these different" until the person answers the question correctly 50% of the time. which is a different question than "where, precisely, is the line between these two colors". with the different question, it ends up compressing the result down by about an order of magnitude.
absolutely! thanks for posting it and the associated article.
To promote some further reading:
OKLab isn't actually a perceptually uniform colorspace. It's better than others, but it was specifically chosen as a tradeoff between accuracy and speed (hence the name OK). When you start digging this deep, you quickly learn that we have yet to invent any perceptually uniform colorspaces; even the most precise models we have end up using fits and approximations. Color has some really inconvenient properties like depending strongly on brightness and background. Frankly, given the differences in human biology (having orders of magnitude differences in relative numbers of each cone, for instance), it's surprising we agree as much as we do! Human color perception is an endless pit of complexity.
(Note, I don't say any of this to detract from what you've built here, merely expand. Your site is awesome and I love it!)
Had to turn off the "Night Light" (reducing blue) and set brightness to max.
- 0.0028 on my MacBook pro screen
- 0.0045 on my Dell monitor
- 0.0033 on my Pixel 10 pro
And those scores are pretty consistent.
it is measuring the smallest color distance you can still detect. so a lower number means you can spot the difference between two more-alike colors.
This reminds me that there's a worlde like game for color mixing, I think someone posted it on HN a while ago colorfle.com
Very addictive, kudos to the dev