Slashdot Mirror
Computer Scientists Scour Your Holiday Photos
Barence writes "Hundreds of thousands of images on Flickr are being used to teach a program to determine the geographic location of an image, simply by looking at it. The program attempts to mimic the way that humans can deduce the location of an image by searching for visual clues, such as similarities to pictures or locations they have seen previously. In its current state it can guess the location of a photo to within 200km, 16% of the time — extremely accurate given the complexity of the problem."
16% percent of the time it works every time.
The paper referenced in the article has an interesting density map of where their 20 million source photos were taken (ok, so they only ended up using 200 or so of these). It says it uses a logarithmic scale, and seems to imply that the vast majority of photos available to them on Flickr were taken in one of only a handful of locations:
Ok so there are a couple more than this, and my geography is appalling, but these seem to be the only areas that are are coloured red.
Dude, that's as accurate as my girlfriends map navigation. *sigh*
War is the statesman's game, the priest's delight, the lawyer's jest, the hired assassin's trade.- Shelley
I'll guess...New York City, without even looking at the pictures that should get me in that ballpark.
Have you read my blog lately?
... is a program that will remember the names of the people in the photos.
Intron: the portion of DNA which expresses nothing useful.
Really? But I didn't go anywhere on holiday. How can you locate that?
Show it a picture of the andromeda galaxy and throw its statistics way off.
If you can read this, I forgot to post anonymously.
As an afterthought did Cheyney and Rumsfield use a beta of this a couple of years ago to find something?
War is the statesman's game, the priest's delight, the lawyer's jest, the hired assassin's trade.- Shelley
extremely accuratean interesting start.
Nerd rage is the funniest rage.
I'll no longer have to spray paint "I was here" for people to know where I was.
of the time...
(Not counting those rich bastards who can afford taking a holiday on the ISS).
The headline looks like a LOLCATS title
"Computer Scientists r in yur webpagez skowering yur fotos"
Would they say the geographic location of Goatse to be... Uranus?
Where's Goatse?
Some drink at the fountain of knowledge. Others just gargle.
Well, there's spam egg sausage and spam, that's not got much spam in it.
metacafe link here and TED link here.
A-Bomb
Scientists surprised to discover it is possible for a machine to loose will to live.
1/2 of the circumference of the earth is just over 20,000 km, so you are within that distance by surface travel 100% of the time.
OK, maybe you meant the through-the-earth distance. At the poles, that's 6,356.8 * 2 = 12,713.6 km. At the equator, it's slightly more.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
It's a for loop that spits out "Your mom's basement".
Look at this set of pictures:
http://htmlhelp.com/~liam/Hawaii/Kauai/WaimeaCanyon/
Would you know simply by looking at the photos without the sign that this was not say the grand canyon? The whole correct to 200 km aspect is troublesome when the state of the art in computer vision cannot yet even answer that this is a picture of a canyon.
OsamaBinLaden2001 has deleted his account
200km, 16% of the time? I guess that sounds sorta neat... except that 84% of the time, it's off by more than 200km. Now, we know that the earth's circumference is 40000km, and it follows that nobody can ever be more than 20000km from any location on Earth.
So 16% of the time, it's accurate to within one percent of the TOTAL RANGE OF ERROR. The other 84% of the time, you're on your own. I wonder if I could manage that kind of accuracy just by sampling colors, classifying them by terrain, and then just picking a likely spot at random.
Yahoo! Pipes are awesome. How awesome? http://pipes.yahoo.com/jesdynf/slashdot
I only skimmed the article but at least we are getting some good scientific use out of all the social networking gobbledegook we have floating around out here.
From the looks of the test selecting London all the time would have a
1/6 chance = 16.67% chance.
They need better double blind testing and a more diverse set of geographical locations.
...by lots of photos of amateur, um, "naturist" pics?
"Hey, Frank? Why are there giant palm trees in Washington D.C.? And why is the Washington Monument pink no...
Oh, never mind."
Yes, I meant through the Earth distance and yes I did manage to use the radius.
That will teach me to post before drinking my coffee...
mcgrew's razor: Never attribute to stupidity that which can be explained by greedy self-interest
Surface Area of Earth
510,065,600 km2
*.33 = 170021866 \\Estimate 2/3 of earth is ocean
/ 3.14 * 200^2
=8%
I propose we all take pictures with blue screen in them (not the whole background, just "enough") and then write a script to randomly replace the blue screen with alternative locations every time the picture loads.
It will reply: "Nice try! That's a reproduction of the Eiffel tower glued on a 1000/1 model of London, in Central Park!".
Seriously, while we have made great steps when it comes to brute force calculations, projects like this one still need that bit of artificial intelligence; something we're still in its early steps of development.
Unless your at that tacky "NYNY" casino in Vegas.
Ubiquitously - A Ubiquity Developer Community
I hope they can extend this to identifying the location down to the nearest street. It's possible to do this if there are some obvious hints like a postcode on a street nameplate. Having a webpage address or telephone number on a shop display can help too. Even a original shop name, or a unique combination of high street stores side-by-side can help.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
The problem with geo-tagged flickr photos is that in many places the detail on the maps and aerial shots provided isn't defined enough to allow an accurate placement.
The even bigger issue is that, although some cameras now have GPS, the majority of geo-tagged shots are placed manually by humans who often get it wrong or deliberately place their photos onto a more popular location just to increase their traffic.
I'd like to present this with Moon landing pictures to see where the moon landing was staged! (hahaha... love it)
In its current state it can guess the location of a photo to within 200km, 16% of the time â" extremely accurate given the complexity of the problem."
60% of the time, it works every time. They've tested it.
I drank what? -- Socrates
I, for one, welcome our new photo-scouring, location-pinpointing overlords.
Use this awesome technology to find Bin Laden and other terrorists who likes to videotape whenever they feel like 76 virgins are any near. 200km 16% of the time! hell yeah. lets invade Iran now! Oh wait..
How long before some one says "I submitted a photo of my girlfriend, and it suggested that she looks like Idaho"
1. Humungously huge person who is so big they cannot possible obey the laws of motion and momentum - must be the USA.
2. Miserable looking person burning an English language book - must be France.
3. Happy looking tanned person laughing at miserable person in 2. - must be an Italian football fan.
Oh, and because someone will do it if I don't:
4. Person with bad teeth and lying drunk and comatose in the gutter - must be the UK.
Gentoo Linux - another day, another USE flag.
You could probably get within 200km better than 16% of the time if you always guess the photographer's home city.
I also wonder how well you could do by checking the photo's timestamp, then examining the shadows to determine the sun angle.
I once correctly guessed "Quebec" in a National Geography Bee when asked in which Canadian province a picture of a particular attraction lay. My clue? The sign on the front of the bus was in French.
Ceci n'est pas une signature.
If there are 6 groups, then the chance of randomly guessing the correct group is 16.66%, not 2.75%.
Let's say we pick the same city every time.
We have a picture. It's from one of 6 locales. We pick a locale C. What are your chances that you're right? 1 in 6.
We have another picture, and pick locale C again - chances? 1 in 6.
Another picture, we pick locale C, and again, chances are 1 in 6.
You seem to understand this concept when you say the chances are 1 in 6 when we pick the same location each time.
So, let's get adventurous, and the next picture we look at, we're going to guess A.
What are the chances that the picture we are looking at is from location A?
ONE IN SIX!
It's 16.66% no matter what.
paintball
The dice analogy is right-on.
The problem is he just doesn't seem to realize that the chances of throwing doubles are 16.66%.
paintball
Make sure your coffeemaker uses a good filter.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
It requires some sort of seed (GPS data, location name) - so such an input set would be able to narrow down possible (known) locations for those photos. It's not going to give you a location result for somewhere it doesn't know about (yet).
16% of the time, it works every time
Reminds me of the experiment done in a Dutch military lab a couple of years ago. They trained a neural network to recognize whether a photograph taken out on a country road had a military vehicle in it or not.
The system recognized the photos from the training set perfectly, but did no better than random on images fed to it that were taken at different times.
Turns out all the training shots with a military vehicle in it had been taken on a sunny day, and the control shots without one had been taken when it was overcast. The system had been trained to recognize a different thing from what they intended!
Do you think it helps or hurts that my photos on Flickr have titles like "Tokyo - Ueno park"?
If you do what you always did, you get what you always got.
6,356.75000 * exactly 2 = 12,713.50000 or 12,713.5000 give or take.
6,356.80000 * exactly 2 = 12,713.60000 or 12,713.6000 give or take.
6,356.84999 * exactly 2 = 12,713.69998 or 12,713.7000 give or take.
If our initial radius measurement of 6,356.8 is accurate +/- 0.5 km, then the diameter is between 12,713.5 and 12713.7 km. We use 12,713, by convention, even though we know we are much more precise than the +/- 0.5 km implied.
On the other hand,
6,356.8 + 6,356.8 = 12,713.6 by convention, even though we could be off by as much as 0.1km either way.
Take your pick.
The problem is, the diameter of the earth at a given point through the center of the Earth may not be twice the distance from that point to the Earth's center.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
because that's the radius of the earth, so the center is the only point where your claim is true
:-)
HAHA
There's an upcoming paper coming from MIT on this topic, Recognition of Natural Scenes from Global Properties: Seeing the Forest Without Representing the Trees that proves this isn't as hard as you might think.
To sum up this massive paper in a very small (and likely highly imprecise) nutshell, building models up from basic objects (the traditional method) is only one way to approach this. Using this method, you are correct; it's impossible to understand what a canyon is. Using the new global properties methods in this upcoming paper, you can gather basic elements that could easily help in assigning location properties, understanding that something pictured is a desert or forest, and theoretically using that data to help determine which desert or forest (this latter portion is beyond the scope of the paper, but great fodder for a future paper that builds upon these fundamentals).
While the method currently requires a high level of labeling in its images, it is hoped that this labeling becomes unnecessary on larger data sets.
Use my userscript to add story images to Slashdot. There's no going back.
That's the "joy" of artificial neural networks (ANNs) - they will focus on the things that change and will learn to ignore the "noise". In your example, the lighting was the "key" that was consistant in the two types of pictures.
My research involved ANNs (specifically, Self-Organizing Maps, but I've worked with the more "traditional" feed-forward-back-prop nets I suspect you were discussing here). From my own experience, what the parent post said is quite probable.
I guess the folks in the Dutch military didn't understand ANNs well enough to guess that. The "ideal" set of photos would have been pictures taken in sequence: take a photo, then drive the vehicle into the scene and take it again, then take a third with a non-military vehicle to help it learn the difference between military and non-military. It can work. I'd use a Self-Organizing Map, though, because it groups things into categories instead of simply giving a "yes/no" answer.
Is it in sample performance ( using the training set ) or out of sample performance ( using unseen images ) ?
How were the picture selected ? Pick the geographical center of the US east coast each time and you should get a decent result for example⦠TFA says it's 20 times better by random, but do they mean purely random (New York and the middle of the Pacific equally weighted) or random based on the distribution of the geographical locations of the set ?
What is the distribution of the results ? It is sometime very accurate, sometime not at all ? Does it generally gets the right continent or does the performance merely reflects easy hits (oh, the Statue of Liberty, oh the Eiffel Tower, etc)
Read the paper, the article is completely uninformative.
\u262D = \u5350
Google should get behind this. I think their Picasa would benefit from it.
Generate some autotags.
What would be nice also is if they had a feature where if you labeled someone in a picture, if you uploaded another picture with that person in the picture, the program would prompt to auto tag.
I've been going through old family photos and it would save so much time if the programs I am using autolabeled based off details in the picture.
"Only one thing, is impossible for god: to find any sense in any copyright law on the planet." Mark Twain
They have dueling Starbucks on some street corners... now you have to guess which one it is!
Comment removed based on user account deletion
Comment removed based on user account deletion
This is the basis of Limbo of the Lost. I'll be here all week.
Does it cue on the date, time, and sunlight conditions? Because that's how I would write it. A computer could easily narrow the geographical location of a photo that way, but humans don't really do that without a conscious effort.
If I post a few photos from a recent endoscopy, can this program guess where I had it done?
Or, at least, tell me if my piles are recovering?
Backward%20compatibility%20is%20over-rated
16% percent of the time it works every time.
College-Pages.com - Online Colleges, Degrees, and Programs