faetal on 18/4/2015 at 20:37
It still looks blue and black to some 28% of people (myself included). What is happening is that if you wash out that particular blue and black, then past a point, the colour values will be interpreted, by way of an optical illusion, as white and gold. However, due to there being an overall variation in the exact proportions and intensity thresholds of people's cone cells, as well as cortical processing variances, a percentage of people will see white and gold and the rest will see blue and black (excepting persons with chromatic visual disabilities). Factor in the variance in monitor brightness as well (which explains why when I saw it on my laptop, it was white and gold until I "reminded" by brain when I tilted the screen).
The interesting question I guess is not whether the dress is black an blue (which it is), but whether in the picture is appears black and blue. This is interesting because whether or not the actual colour values (depending on the screen) are objectively closer to black and blue or white and gold can only be done by looking at the pixel values. Because what is more likely, that an actual blue and black thing is providing an *illusion* of being white and gold, or that it is providing an illusion of its original colours, when accidentally faded to different ones? Ockham's razor and that.
WingedKagouti on 18/4/2015 at 21:32
Quote Posted by faetal
The interesting question I guess is not whether the dress
is black an blue (which it is), but whether in the picture is
appears black and blue. This is interesting because whether or not the actual colour values (depending on the screen) are objectively closer to black and blue or white and gold can only be done by looking at the pixel values.
The top part is R70, G50 & B30, which is darker than the darkest part of the dress I could find in the image in the OP. The bottom part is from a color pick of a random spot on the blue (which is easy to mistake for white in that image) part of the dress.
Inline Image:
http://i.imgur.com/d0wBxr1.pngbjack on 18/4/2015 at 21:33
Funny thing. I usually use my Mac Book to surf the internet and have looked at this dress thing only on that bit of kit. I'm now on my old XP box and looking at a 27" screen at a 30 degree angle upward. The dress in the posted picture now looks goldish brown (almost burnt sienna) and pale bluish silver. Since my screen is much larger than the Mac, as I scroll up and down the color fades from dark to light (as it does with everything btw).
I did not want to break any copywrite laws, so I did not post a pic of the real dress color, only the link to the dress. Has anyone looked at the link? If you look at the real dress, you will see that the posted pic here is nothing like the real dress in color. Turmeric/mustard turns into absolute black. The semi bluish purply white turns into a bright medium blue.
Muzman on 18/4/2015 at 23:30
...and the next thing that happens in these arguments is people start saying, in effect, that even though they were wrong they were right. Or that they were more right to be wrong. That their wrongness was closer to right, or more conducive to rightness in most situations. It's the right that were actually wrong, because their rightness was arrived at by a fluke, or error or crazy eyeballs. Wrongness just makes the most sense! Wrong is reasonable.
Which is great, I love it.
So anyway, a fun thing to do is remove the colour information and then look at it
Inline Image:
http://members.iinet.net.au/~ragtag/dress1bw.jpgIt may or may not make any difference, I suppose. But it should make it more apparent that the lighter of the two materials is in no way close to white and barely rises out of the midtones of the image.
More interesting still is to pay attention to the
other white highlighted dress behind it.
Inline Image:
http://members.iinet.net.au/~ragtag/dress1crop.jpgIf you look at the black part of that material and check the values for the lighter and more illuminated parts of the folds, the values you get are between 25% and over 45%, or about 64 to 115 out of 255 with a rough typical value of 76 or 30% (not scientifically attained, just the value the colour picker seemed to find the most)
If you look at the black or 'gold' part of the dress we all love to hate, you get values of around 20% to 45% or 51 to 115 out of 255. The typical value is also about 30-33% or about 76-84 on a 255 scale. The brightest part at the top, where most people seem to get their gold impression from, is only 45%/115 at the most.
Now looking at the white part of the other dress, behind the main one. As deep as the image resolution will allow in the folds there, the white part is 75%/191. In the main it is over 95%/242.
Now that, folks, is white.
The blue/'white part of the controversy dress is typically around 55-60% or 140-153. At its tippy top highlight up on those shoulder folds you get 83%/212. Far above the average for that material in the image and nowhere close to the other white, on average.
So we have two black materials which, when illuminated, report roughly similar tonal values, as expected, even though one looks gold to some. And two purportedly white materials which are wildly different.
But, they say, why measure the high point of the other black against the darker material of the main dress?
Because that's what's happening here. At first glance I read that the main dress is quite well front/top lit (and the whole image is probably brightened again after the fact). It's not particularly bright lighting compared to the exterior, but nothing can compete with that. That interior lighting, coupled with a shiny material and some post brightening are going to give you quite a high average for that sort of black material. And moreover, they would give you very obvious highlight whites, as the other dress shows. Ergo, it is not white. It is, as it appears, blue. (and one would guess quite a dark blue, under true-er conditions. Which the real thing bears out).
One remaining question is, what role did this reason and experience play in my actual viewing of the picture the very first time? That's the hard part, really. I saw blue/black right away and I can tell you why, but did it make any difference? I definitely have seen plenty of people who have no idea what they're talking about who saw blue and black that first time as well (and it's debatable that I know anyway). So who the heck knows why.
I imagine some psych students are running around studying it right now to see if it's a 50/50 split or there's some other predictor. All kinds of fun anyway.
van HellSing on 18/4/2015 at 23:37
Im' drunk and I just saw it as blue and gold. I'm outta here.
faetal on 18/4/2015 at 23:50
The images you just posted are particularly desaturated versus the ones in the OP.
What's even more interesting is how people get annoyed / passionate about their particular interpretation of the colours. I wish I knew enough about psychology to be sure about why that is.
Muzman on 19/4/2015 at 00:02
Quote Posted by faetal
The images you just posted are particularly desaturated versus the ones in the OP.
Not particularly. Completely ;)
It's funny, and perhaps distracting, that around the same time this was going around there was some coverage of that study which showed that when people are shown their views or impressions are factually wrong their first reaction is usually to dig in or double down and come up with some convoluted was of saying that they're still actually right in some way.
After reading that story I interpreted these sorts of debates through that
a lot.
faetal on 19/4/2015 at 00:08
The fact that people see different things is very interesting.
Once figuring out it's an illusion, less so. Then realising that most people see the illusion, more interesting. Then realising why, less so. Then seeing people on facebook and twitter *insulting* each other about it, very much more so.
Seriously, where the psychologists at?
demagogue on 19/4/2015 at 22:53
I read a cogsci article on color perception that was looking at classic art where artists used white paint for clearly black objects, like (
http://upload.wikimedia.org/wikipedia/commons/c/c0/Gustave_Caillebotte_-_The_Floor_Planers_-_Google_Art_Project.jpg) floor1 (bottom center bit), (
http://www.thefamousartists.com/wp-content/uploads/2010/10/The-Washer-Women-v1.jpg) floor2, and (
http://4.bp.blogspot.com/-a01fyDMDwAM/TcqyVIqL0-I/AAAAAAAAAAk/xtG_p7Ew4Bg/s1600/another%2B%2Bboo.png) floor3. Josef Albers is associated with modern art theory of it.
The cogsci article said shadow and 'base color'get proccessed in parallel, so that we can't see the white paint as white, but black. Even when the floor1 black strip is next to a 'darker' peeled stripe, we can still tell the black part in white paint is darker. The process here is probably then whether the black in the dress gets put on that 'shadow coloring' track or not. If it does, it'll get processed black, if not then you get the white. Edit: Sorry, vice versa of this. It's not 'shadow coloring', it's 'lighted coloring', so if it's on that track the off-white appears black, and if not the off-white looks white.
Here's my first theory I'll leave for completeness sake--
There's probably multiple things going on at multiple levels. I remember from cogsci the effect of illusory shading to accentuate boundaries, or part of it, happens very low level in the ganglions just behind the retina already. But for things that 'pop' in or out, like the rabbit-duck family of illusions, there's an interpretation of the overall context or situation that puts a feature down one perceptual track or another, and fills out the details accordingly. That's why they call perception heterachical, because it's top down (context directs features) and bottom up (features direct context) at the same time.
zombe on 20/4/2015 at 12:03
Quote Posted by demagogue
I remember from cogsci the effect of illusory shading to accentuate boundaries, or part of it, happens very low level in the ganglions just behind the retina already.
Some quite important conversions and processing happens in the retina - however, none of them are relevant in the dress context. As far as i understand it of course.
In essence, the center-surround / color-opposition for cones can be seen as a very small kernel sharpening convolution filter ... way too small to matter. The net effect is just color-opposition conversion.
Similar sharpening effect is applied to rods too, but on a larger scale (*) - however, it does not code color.
*) it was once thought to explain (
http://en.wikipedia.org/wiki/Grid_illusion) grid-illusion ,but having gone through (
https://www.youtube.com/playlist?list=PLUl4u3cNGP63Wy1oqQW2oD2HBdDSY8Ghi) MIT OpenCourseWare - Sensory Systems, Fall 2013 i have to agree that it does not. Wiki notes: "There is strong evidence that the retinal ganglion cell theory is untenable. /.../"
--------------------
My take on the dress.
1. Color classification as in "is this brown?". There are only 4 pure colors - the non-mixed points of red/green and blue/yellow opposition. You can reasonably expect (given no color deficiencies) thous edge cases to be classified most consistently - but the rest is considerably affected by cultural and personal background.
Most common neural structures contain an element of vertical paths and its accompanying lateral inhibitions which, in essence, is a classifier (*). One fairly obvious side effect of such a classifier is strong narrowing of the middle-case (ie. cannot decide which) by widely broadening the bordering cases. Thous middle and bordering cases are learned and do not match culture-to-culture / person-to-person. => One says it is certainly color X and certainly not color Y while the other says it is certainly color Y and certainly not color X not realizing that their brain did some classification for them and their perception of it overplayed its certainty.
ie.
Code:
XX??????YY more objective classification
XXX?YYYYYY one person's bias
XXXXXX?YYY other person's bias
obj-^
(*) retina has also this structure in its purest form - ie. sharpening convolution filter itself is a classifier (... of _significance_, whatever that significance may be. Since the input for retina is just plain contiguous luminance then the significance is the change of it -> ie. edge detect [sharpening filters and screen space ambient occlusion filters in games are just glorified edge detection filters]).
2. shadow/ambient color correction as in "(
http://serendip.brynmawr.edu/bb/contrastcolor/cube.jpg) is it brown given its surroundings?". Good old classifier (lateral inhibition in various cortical columns [however you define the term]) doing its work again.
Abundance of color X in one classified region suppresses itself reinforcing the outliers there to shine through. Does not do much with the 4 pure colors, but can have quite dramatic effects with the mixtures (especially given point 1). Nothing functionally new here.
Since we cannot see much of the surroundings of where the dress is then the color correction is more wrong than it normally would be and is highly dependent on the unrelated light conditions around it (other shit on screen / your desk [and the cat asking for food on it] / your room)
3. Actual color as in "brown paper is brown". It is technically strictly impossible to extract the actual color of the dress from the picture. You need the dress itself to inspect in a controlled/known light conditions.
Add thous three together and you get a constantly shifting mess of confusions and opinions to fight over.
------------
/.../. Apparently the food cannot wait.
