Computers With Opinions On Visual Aesthetics 125
photoenthusiast writes "Penn State researchers launched a new online photo-rating system, code named Acquine (Aesthetic Quality Inference Engine), for automatically determining the aesthetic value of a photo. Users can upload their own photographs for an instant Acquine rating, a score from zero to 100. The system learns to associate extracted visual characteristics with the way humans rate photos based on a lot of previously-rated photographs. It is designed for color natural photographic pictures. Technical publications reveal how Acquine works."
Computers with opinions?! (Score:5, Funny)
Re:My lone opinion (Score:5, Insightful)
The field of AI is not comprised of a majority of researchers frantically trying to build an expert system that can pass a Turing Test. Visual data is complicated and building a system that can take that information and make use it in a very simplistic manner [wikipedia.org] is non-trivial. Read some [stanford.edu] of scientific papers [stanford.edu] published by [stanford.edu] the authors of Acquine, and you'll see that their methodology (image processing, regression, Bayes' classification, decision trees, support vector machines, classification and regression trees, to name but a few) is anything but trivial.
Not only did they build something novel, but they built a system that does a good job of approximating human response to good/bad photography.
If you want to contest the true novelty of their work, through an academically-inspired claim that they combined existing technologies in a way that isn't terribly novel, rather than creating their own technology, then that's fine. However, the blanket statement that some researchers are trying to do "real work" [robocup.org] and that Acquine isn't real work, is a giant red-flag indicating that you likely haven't got the slightest clue what actually goes on in the field of AI. Typically researchers like to tackle problems where the utility of their solutions isn't immediately obvious, the previous link to the RoboCup competition is a perfect example; who cares if you can build a robot that can play soccer? By your reasoning, that would be an incredible waste of time. Except, it's becoming the standard problem for multi-robotic systems research, and a large number of AI researchers are devoting significant time towards building RoboCup teams.
Why?
Simple, pick a real world task for a group of robots that "matters". Now decompose that task into all the subproblems that you would need to solve in order to have a robot complete the main task. Chances are, you're going to run into problems involving self-localization, team-work/cooperation, vision, data fusion, etc... All of those subproblems are being worked on and solved in different ways by researchers in the RoboCup challenge. And chances are, if you choose the methodologies used by the teams that win games you're likely to have chosen the most effective methodologies available in the field.
The true value of research isn't the end-product of each individual research-project. It's the end-product of many "useless" research-projects combined.
Re: (Score:1)
Re: (Score:2, Informative)
Do you find that it "does a good job of approximating human response"? Because I don't. Photos that are absolute crap get 99.99% and interesting photos get 1%. It seems completely random, except for its apparent preference for high resolution and black borders.
Re: (Score:2, Funny)
Re: (Score:2)
Sadly, it didn't take long for me to run into this image while using acquine in random browsing mode. It got 4 out of 5. So there you have it: goatse is art!
No, I am not making this up, either.
Re: (Score:2)
The field of AI is not comprised of a majority of researchers frantically trying to build an expert system that can pass a Turing Test. Visual data is complicated and building a system that can take that information and make use it in a very simplistic manner is non-trivial
Chances are that the types of algorithms it takes to make sense of incredibly complex visual data are the same types of algorithms you would need to pass a Turing test.
The dangers of trusting machines... (Score:3, Insightful)
So I follow the link and think, hey this is pretty cool. I grab some screen shots of my apps and run them through. Unsurprisingly, some old VB6 crap I'm still maintaining was scoring in the 5-10% range. The newer Web and Silverlight apps I've been working on are all over from 30%-70%. I'm thinking this software is pretty cool and we could use it to get rapid feedback on different layouts and styles.
So I send the link to one of my co-workers. He brings it up and posts a screen shot of his web site. We start
Re: (Score:2)
Signed: Redundant array of inexpensive employees, high availability node #145738433
Re: (Score:2)
Funny, eh? I, however, don't need to know the visual aesthetics of the goat.cx dude, which is what I was greeted with when I went to the actual site. Thanks, it was the perfect troll.
Re:Computers with opinions?! (Score:5, Funny)
Oh, this takes me back to something that happened when I was still at school, around about 1996 or so.
At the time, despite being pretty much clueless by slashdot standards, I functioned (and still do function, when I can't avoid it) as the all purpose IT helpdesk for my family.
Now, I have an aunt who was working as a manager in a medium sized UK based IT firm (I can't remember the name and I don't even think they're still around). However, do not take this as any indication that she knew anything at all about computers. She didn't. Nor did she have any inclination to learn. She could just about manage to use Microsoft Office and the web/e-mail clients that were around at the time (I think this was around the time of Netscape 2). Literally anything beyond that would baffle her. She'd call her monitor "the computer" and so on. She was employed, I gather, for her "management skills".
Now, even at the time, I had a feeling that this was a crock. My aunt is a rather forceful personality. A less diplomatic person might use more bovine terminology and I always got the impression that she wouldn't be much fun to work for.
She's also very, very, very large (over 25 stone) and ugly as sin to boot. That's not me being deliberately rude. There's just no nicer way to put it without doing a genuine dis-service to the truth.
Anyway, one Saturday afternoon, I get a call from her. Her voice implies that she's perched in that dangerous territory between bursting into tears and throwing a screaming, PC-destroying fit. Apparently, her computer is "insulting" her. I need to go over there instantly. I got a lot of this kind of stuff until, a couple of years later, I finally told her where she could stick it after I moved off to university and got a call asking me to travel 200 miles to fix something. Anyway, I'm not best pleased about losing a Saturday afternoon, especially with exams coming up, but for the sake of a quiet life, I head over.
Oh boy was it ever worth it.
Sure enough, every two minutes on the dot, her PC is insulting her. Whatever she's doing in Windows, a little dialogue box will pop up with a splendidly vicious insult. I mean, some these were absolute gems and were clearly aimed right at her personally. A few of the more repeatable examples (and I still remember these more than a decade later) were:
"Careful! Better fetch an extra chair. I think those two are about to give way."
"Wow you must be constipated. Or does your face just look like that normally?"
"Did you just fart, or do you always smell like that?"
"Wipe your face. Half your lunch is stuck between your fifth and sixth chins."
"Is that your face or your arse I can see? Your face? Hmm... the arse might be better."
"I can access over 64,000,000 images via the Internet and none of them are as ugly as you."
"Shall I compare thee to a summer's day? No? How about a half-digested turnip?"
There was plenty of other stuff as well, including the old classics about ID10T and PEBCAK errors, but enough of it was specific enough to my aunt (making mention of particularly distinctive unflattering features) that this was clearly something bespoke.
Anyway, my aunt's in an absolute state at this point. She's convinced that the computer is insulting her. She tells me she tried covering up the monitor for half an hour (so it couldn't see her), but when she came back, it had been queuing up the insults.
Anyway, having confirmed that a virus scanner doesn't pick anything up, I ask to see any disks she's put in the PC lately, or any files she's downloaded. The downloaded files all look pretty safe, and it doesn't seem like anything dodgy's come in via e-mail either. However, she then shows me a couple of disks (3.5" floppies) she'd brought home with work on. These are a numbered series of progress reports. Most of the disks look absolutely fine - a few Word and Excel files. Nothing too scary (I don't think MS Office files were being used extensively for exploits at the time).
However, on the fi
Re:Computers with opinions?! (Score:4, Funny)
Dude, the algorithm is simple. B&W pictures automatically get 85 points.
Re: (Score:1)
Re: (Score:1)
Computer, play something by Billy Joel (Score:2, Funny)
I'am afraid I cannot do that Dave.
Matter of opinion? (Score:5, Insightful)
I'm not saying "my opinion is better", just that it seems sort of pointless to assign a value to a picture like this.
Re:Matter of opinion? (Score:4, Funny)
My computer says you have no taste.
Re:Matter of opinion? (Score:5, Informative)
Just in case anyone asks, here's the computer rating for goatse, emphasis mine:
Machine Prediction of Aesthetic Value
Score predicted by Acquine: 19.4/100 rated as below average
Looking at that image a
Re:Matter of opinion? (Score:4, Funny)
Tubgirl, on the other hand, rates a 58.8/100.
Someone please stop me.
Re: (Score:2)
Lemonparty: 28.8/100
The goggles! They do nothing!
Re: (Score:2)
Penis Bird: 17.7 / 100
Acquine may be on to something.
Re: (Score:3, Funny)
Acquine's rating of the first Google Image result for "Natalie Portman Hot Grits:"
76.1/100
Judging from my quick research, it's clear that Acquine is going to turn into Skynet someday.
Re:Matter of opinion? (Score:5, Funny)
By the way, I uploaded a screenshot of the Google results for "Slashdot Karma Whore" (all text) and got a 42.4/100.
Re: (Score:2)
71/100
That's right; a red box with a white>black gradient in the background is just that damn impressive.
Re: (Score:1)
Bill Gates laying on desk: 79.4%
In other, unrelated news... (Score:5, Funny)
In other, unrelated news, Penn State researchers have released an online image voting system called 'Aesthetic or not' where users are presented with a random image and have to give it a score of between 0 and 100.
Initial user participation was good until for some completely unknown reason, 90% of images presented to users for rating were goatse, tubgirl, or other shock images.
Re: (Score:2)
Initial user participation was good until for some completely unknown reason, 90% of images presented to users for rating were goatse, tubgirl, or other shock images.
I assume you were joking, but it's actually playing out that way. One out of every four of the images on the front page looks like the type of thing a slashdotter would contribute, and the site says:
5/15/2009 Due to a slashdot exposure, we are experiencing a much larger traffic to our small Webserver.
There is a possibility that your photo request may not be handled as fast as it should be. Thanks!
Goatse got 76.1 (Score:2)
No kidding, I tried it with the url http://www.goatse.fr/hello.jpg [goatse.fr] and got 76.1
I see Slashdot at work here. Well, I also clicked in the rightmost star in the "How do you rate its Aesthetic Quality", of course...
Re: (Score:2)
Huh? I saw it at 4/5 stars on Acquine.
Re: (Score:3, Insightful)
I'm not saying "my opinion is better"
I am. At least they tried something new.
Re: (Score:1)
A third dumb story on /. in mere 24 hours.
Crappy system... (Score:2)
This picture gets 95.1% [www.yle.fi]
This, completely different picture, [abm-enterprises.net] gets 9.8%.
While this one [lairweb.org.nz] gets 7.3%.
Note that the one that actually represents the photographed object the best got horrible grades, but the blurred, oversaturated and cropped version got nearly perfect score.
Rubbish. (Score:4, Funny)
It's terrible. Awful. A hopeless system. I wouldn't use ever it.
And I'm not just saying that because it rated a couple of my photos as poor. :)
Re:Rubbish. (Score:4, Interesting)
It rated one of the best pictures I've ever taken as 13. Then it rated a fairly generic cityscape at around 60.
I think it has some learning to do.
Re: (Score:2)
Well, maybe that means not that it has some learning to do but that your 'greatest photograph' isn't as great as you thought on an absolute scale. After all, aren't you inherently biased?
Re: (Score:2)
Sure, I'm not saying it's actually that great on any sort of absolute scale, but IMHO (and that of others I've shown some of my pics to), it's a good shot and, not being that photographically inclined, those are unusual for me.
But I am saying that I fed it some much more mediocre crap and it loved it. Perhaps the problem it has is inherent - this whole topic is so subjective that it's pointless.
Re: (Score:1)
Re: (Score:2)
Now we just need POQUINE (Score:2)
Re: (Score:2)
Re: (Score:3, Insightful)
That's not how Slashdot really works. Due to the number of astroturfers here, anything critical of Microsoft, no matter how true it is, is usually modded "Overrated", while "Vista works fine for me" gets "Informative".
Re: (Score:1, Offtopic)
Re: (Score:2)
No seriously, exactly how many people do you think Microsoft pays by the hour to browse slashdot and post pro-microsoft stuff? How much would one guy posting a few comments an hour, with no guarantee that he will be modded up, benefit microsoft financially? Seriously? Are you really so out of touch as to think that anyone who posts that Vista is working OK for them in the middle of a traditional MS-bashing is an astroturfer?
Comment removed (Score:4, Insightful)
Acquine may assign funny scores... (Score:5, Insightful)
Re:Acquine may assign funny scores... (Score:4, Interesting)
I noticed this with a picture I took in France that everybody praises. It got 6.9. I did the smallest possible change in color, darkening it imperceptibly. The new version got 35.7. Doing a selective gaussian blur also tends to raise the result a lot.
My rating of their algorithm is 0.01 star, which can be summarized as "it sucks".
Re: (Score:1)
At least you made some colour changes. I handed it the thumbnail version of a photo and got dramatically better results than from the high-res version of the same image...
Re:Acquine may assign funny scores... (Score:4, Funny)
Looks your rating is accurate
Re: (Score:1)
Re: (Score:2)
The image had the best possible quality that would fit in the allowed file size. I believe their algorithm is biased toward portraits, that have a big sharply focused central subject and an out of focus background.
My picture was a boat in a canal, and it had great depth of field. What the selective gaussian blur did was to soften the surrounds, like branches, leaves, and reeds in the image, while keeping the boat in the center in sharp focus.
Re: (Score:3, Funny)
It looks like one would get more consistent results out of the checksum for a picture ;)
Thanks for testing that - not many would have.
What exactly is a "color professional photograph"? (Score:3, Insightful)
What exactly is a "color professional photograph"? Landscapes? Portraits? Group shots? Sports photography? Photo journalism? Abstracts? Artistic Nudes?
This may be an interesting programming toy but it has little to no use in the real world, unless you have a desire to locate generically boring pictures built to formula. (or, generically boring pictures that have been run through the "ALIPR Picture Score Optimizer" Photoshop filter)
terrible (Score:5, Funny)
It prefers a nazi germany flag over some beautiful landscapes and portraits
my name is godwin and I approve this message
Our redundant brains (Score:4, Insightful)
This is great news, a system to tell us whether a photograph is beautiful or not. We are approaching the point where we can outsource all our thinking to computers. Soon we won't have to use our brains at all!
(Not that many of us do presently, anyway.)
Re: (Score:1)
Re: (Score:2, Funny)
Wow, what an uhh... <looks at what mods rated your post>... insightful posting! I'll have to wait to find out what kind of posting mine is.
Re: (Score:2)
Apparently, it's kinda funny but not really.
Re: (Score:2)
I already maintain my car when it tells me to.
Pulitzer versus Goatse.... (Score:5, Funny)
Just google image search for any Pulitzer Prize winning photo and upload it to the Penn State ACQUINE system and see how some of them fare to the Goatse image...
The Iwo Jima flag raising photo at this URL gets a 26.1 in the system.
http://surreality.info/up/WW2_Iwo_Jima_flag_raising.jpg [surreality.info]
The fucking Goatse image with a construction crane photoshopped into it (don't ask) just got an 84.1 on the same ACQUINE system....and no I'm not going to provide a URL just test it yourself.
So Goate is a better image than the Iwo Jima flag raising photo?
Am I missing something?
Re: (Score:1)
Am I missing something?
Slow news day on Slashdot. Did you look at the previous "story"?
Re:Pulitzer versus Goatse.... (Score:5, Insightful)
So Goate is a better image than the Iwo Jima flag raising photo?
Maybe all the people sending goatse to it has biased its aesthetic judgement.
Re:Pulitzer versus Goatse.... (Score:5, Insightful)
How on earth should an algorithm know how to infer the symbolic value of the flag rising image?! As far as I understand the Pulitzer Prize is not about artistic and aesthetic value, but rather about journalistic impact, isn't it?
Re: (Score:2, Interesting)
Well that's exactly the point, isn't it? We're a long way from AI being able to judge artistic merits, because doing so often isn't just a property of the image alone, but is related to what the things in the image represent.
Now having said that, it might be interesting to have some judgement that is not biased by cultural perceptions - one that treats Pulitzer and Goatse on an equal playing field, judging what they look like rather than what they represent. But it's unclear what meaning such an algorithm h
Re: (Score:3, Insightful)
Symbolic appeal is very much different from aesthetic appeal.
I think the problem with Iwo Jima is that it is in black and white and the system is designed to rank color images, according to TFA. However, I think we can see a certain ... similarity between the two pictures. One of the criteria for a composition is how the eye is drawn to a focal point in an image. In the Iwo Jima photo, the mound of the hill is sharpened by the triangular form of the squad and the flagpole, drawing the eye to the flag.
Re: (Score:2)
It's not about symbolic value, it's an extremely good photograph. If it was symbolic value, we'd all be looking at one of the shots of the first flag-raising (which exist). This was the second one, after they got a bigger flag. It got famous because it's absolutely wonderful composition, largely (I assume) by chance.
Re: (Score:2)
I don't know, but putting those two pictures together, you've painted an image in my head where they're planting that flag somewhere else I priorly, innocently, assumed, outside of the picture.
Also, you're missing the aspect of sentimental "beauty", or emotion a picture evoces, this picture calls up alot to the viewer, and more the US population as it plays on patriotism, which lives strong in the US.(not saying anything about that or judging, it's a common observation.). A computer c
Re: (Score:1, Insightful)
I don't give a shit about some yank grunts waving a flag about. You're surprised some system immune to patriotic flag waving bullshit doesn't think too much of it? Perhaps you should stick in an American girl and an Iranian girl and see if it marks down the Iranian for having a president who denies the holocaust happened?
Re: (Score:1)
Re:Pulitzer versus Goatse.... (Score:4, Insightful)
Am I missing something?
Yes.
The system learns the quality of photo, not the abstractions we place upon it.
The photograph, in strict terms of quality alone, is rather poor and achieves an appropriate rating. It cannot measure the value of the image.
Re: (Score:1)
Re: (Score:2)
this may be surprising to you and vinylrecords,
but this app isn't claiming to say squat about "art",
it's claiming to simulate human aesthetic reactions.
(and apparently doing a dubious job of it but at least it's trying)
Re: (Score:1)
Re: (Score:1, Troll)
Re: (Score:1)
Somewhere a Google engineer wonders why the sudden surge of image searches for "goatse crane".
Re: (Score:1)
Re: (Score:1)
Because from an aesthetic standpoint, that particular Iwo Jima flag raising photo has a lot of problems with it. It's very contrasty, the highlights are badly blown, the composition isn't ideal and so forth. The so-called "rules" of photography aren't rules because someone said so, they've demonstrated themselves over time to be things that most people, far more often than not, find appealing in a photograph or image. Things such as the rule of thirds, good exposure, contrasting or complementary elements an
Problematic image? (Score:2)
I tried the system, using a high dynamic range pano that I'm very satisfied with, and which has received a bit of praise from other photographers:
http://norloff.org/pano/mirror5x3medium.jpg [norloff.org]
According to the computer this image is worth 28.1 points.
I guess this means that either the image is a lot worse than I believed, or the rating system has problems with it. :-)
Terje
Re: (Score:1)
That's a great picture.
Re: (Score:2)
Thanks!
To make that mirror panorama image I needed 3 landscape format photos. In each direction I used bracketing to take 5 photos with 0.7 f stops between them, i.e. the total range was about 3 full stops or a factor of 8.
I then optimized a pano consisting of all 15 source images, merging each group of 3 with the same exposure into a separate layer.
Finally I used gradient filters, blending all 5 fixed-exposure panos into a final HDR pano.
Terje
Great... (Score:2, Funny)
hmmm... (Score:1)
Interesting (Score:3, Interesting)
One thing first. There *are* certain esthetic and technical rules / guidelines which are what we could call "objective" in the sense that they are very general. For example, a photograph usually looks better if the composition is balanced, if the 2/3rd (or golden mean) rule is used, the lines in the picture are coherent and lead the eye in the right direction (e.g. towards the subject), if the photograph is correctly exposed, colors matched etc. Of course, some of the greatest photographs break those rules; however, like in many things, you succeed in breaking the rules if you know what you are doing, and you cannot do it very often.
I can imagine that you can come up with an engine that is able to detect how "rule conformant" a given picture is.
However, pure formal esthetic judgement is what we rarely mean when talking about a "good photograph".
There is one main issue that will make it very hard to match our "overall" esthetic sense. Firstly, we are unable to detach the image contents from the "pure form". That means, if we see a worried women holding a child, we cannot just look at that as a composition. Also, we are always considering what we know about the subject. E.g. if we have a photograph of a man standing in water, if the photograph ends just below the place that his legs go into the water, we will have the impression that his legs are cut off, and that there is something wrong about the photograph. Finally, facial expression is immensely important for the perceived esthetics of a photograph.
I did some experimenting -- some of the truly great photographs of our times got rather lousy scores (e.g. Dorothea Lange's famous photograph, but also some color photographs as well), while at the same time rather random shots I did of my sons got even five out of five stars. Well. Maybe it will still be useful to someone to filter out the worse photographs.
j.
Re: (Score:1)
Skewed rating (Score:2)
Apparently, there is a skewed (high) rating for: white or black frame around the picture and black and white photographs esp. portraits.
j.
Existing from more than 20 years (Score:1)
Isn't this exactly the same as the matrix metering [wikipedia.org] introduced in SLR cameras in the eighties? Just instead of calculating recommended exposure you get an estimation.
The principle should be similar - get several thousands of really good pictures that anybody likes. The more, the better. Run analysis for patterns and store the results in a database. And when you need to evaluate a picture, just search for available patterns.
Opinion from Penn State (Score:1)
Sunset detector (Score:1)
Acquine is an Opus Dei!! (Score:1)
Oblig... (Score:2)
I wonder what scores it gives to porn?
Oblig. (Score:2)
I, for one, do not welcome our new aesthetically aware computer overlords. I get enough of that from the chicks at my uni.
perhaps it's a trick (Score:2)
and this isn't a computer research project at all but rather a sly psychological study trying to gather a collection of images which real humans are aesthetically interested in. ..perhaps correlated by IP / physical region or perhaps correlated to before and after being posted on slashdot.
nondeterministic (Score:2)
try uploading this image [alipr.com] (it's clean) several times. it generally gets a 98.9, but sometimes gets a 5.6. after mirroring the image [alipr.com] it generally gets a slightly reduced 98.4 but occasionally gets a 35.4.
Re: (Score:2)
it would be interesting to hook this up in a feedback loop to something that randomly/genetically mutated images and used acquine's score as the fitness function.
At last: Conformity (Score:1)
The computer's opinion (Score:2)
Are we sure they are not using a Mechanical Turk [wikipedia.org]?
Of course, of course.. (Score:2)
Beta testers noted that images of mares were particularly appealing to the software.
Inherently conservative (potentially boring) (Score:2)
1) Aesthetic according to what standard?
2) Art is more than photos. Much art moved away from realism 150 years ago.
Some short thoughts [mit.edu] on the topic by a physicist.
Re: (Score:2)
Re: (Score:2)
In fairness, if you RTFA you'd know that they specifically discourage the use of their system to evaluate paintings and hand-drawn artwork; the system is instead meant to evaluate color photographs.