Software Describes Surveillance Footage In AI-Generated Text
holy_calamity writes "A computer vision research group at UCLA has put together a system that watches surveillance footage and generates a text description of the events in real time. It only works on traffic cameras for now but demonstrates how sophisticated computer vision is becoming. Interestingly, the system was built thanks to a database of millions of human-labeled images put together by Chinese workers."
It won't impress me until it can say "Check out the hooters on that chick!"
See my journal for slashdot ID's by year. Mine created in 2005. http://slashdot.org/journal/289875/slashdot-ids-by-year
There needs to be an expectation of privacy regarding recordings of people in public places. There is a huge difference between being seen vs. having one's every public move recorded, indexed and archived.
the system was built thanks to a database of millions of human-labeled images put together by Chinese workers.
I spent a brief amount of time checking out Amazon's Mechanical Turk, and this was one of the activities they offered pennies on the hour for. Yay for crowd-sourced globalization! 100 years from now, when many of the mundanities of life are automated, is this what minimum wage workers will be doing?
'We are trying to prove ourselves wrong as quickly as possible, because only in that way can we find progress.' RPF
Note some of the text in the sample video "Possible Yield violation by Landcar_XXXXX". Are we seriously going to leave policing to an AI? /shudders
"Please describe the scientific nature of the 'whammy'" - Agent Scully
To rid the world of every shred of privacy remaining (not that there is much, admittedly). /shudder
Remember to maintain your supply of
A sophisticated computer vision system relies on a database of millions of human-labeled images put together by Chinese workers??? Reply: RTFM, but dude, can the poster at least make the headline coherent?
This has huge potential to not only push computer vision forward, but also humor.
Example text:
"I see that one old man hobbling down the street, I think he may be off his meds. Uh oh, he's looking _crazier_ than usual!"
"Some asshat just drove completely through a red light. I don't even think she saw the thing! License plate #45AhfD... Is Mrs Doris Johnson-Johnson.. seriously? Who hyphenates the same name!? Seriously I can't comprehend that. But I digress. Her address is .."
The possibilities are endless.
They've added "find similar" links under some pictures. I presume this was an expansion of their "goggles" program on Droid-phones. That was supposed to help locate you by taking a photo of a distinctive object in your vicinity.
Students and geeks have no empathy any more. They're no longer the elite 10%, but tools in extended training to prepare themselves for debt, servitude and an unsatisfying family. Cooperation with China in academia IME is not to improve local academic talent with the best Chinese minds but because Chinese coming to the West are the sons and daughters of rich, well-connected families who pay full fees and more to the Universities concerned (this is even worse in the EU than US, where some Universities are dying for cash and it is official policy to send senior staff over to China to court students).
You can mod down my posts as much as you want because you are too ashamed to admit there might be a problem with a cooperation between US and Chinese academia+government on improving artifically intelligent surveillance tools, but it won't stop the fruits of their labours being used to watch your movements within a decade or two.
Booth: Gun. Noun. Portable firearm. This device was widely utilized in the urban wars of the late twentieth century. Referred to as a pistol, a piece...
Simon Phoenix: Look I don't need a history lesson! C'mon, HAL, where are the god damn guns?
Moral Statute Machine: You are fined one credit for a violation of the Verbal Morality Statute.
Simon Phoenix: What? F*** you!
Moral Statute Machine: Your repeated violation of the Verbal Morality Statute has caused me to notify the San Angeles Police Department. Please remain where you are for your reprimand.
Simon Phoenix: Yeah, right.
[police sirens approach]
Simon Phoenix: F***ers are fast too.
Moral Statute Machine: You are fined one credit for a violation of the Verbal Morality Statute.
Cool. Now the only thing they've got to add is a rating system and I can outsource going to the movies.
Bert
Surveillance Software Knows What a Camera Sees
Cameras don't see, and computers don't know, and anybody who knows anything at all about seeing and knowing and how cameras and computers work know this.
I wish I hadn't clicked the link; I'm not going to read a FA by someone so clueless.
Free Martian Whores!
>> the system was built thanks to a database of millions of human-labeled images put together by Chinese workers
Happy car clash into barrier of non-moving.
Point it towards other surveillance and see what it thinks is going on: "Car 1 is climbing on car 2 and repeatedly rearending it!"
I would expect someone out there to try to do creative things just to make the AI Generated Text say something really weird. At least... that's what I'd want to do. I like challenging and breaking software though.
10 PRINT "A car just went past."
20 GOTO 10
GCHQ Quantum Insert installed. If only our tongues were made of glass, how much more careful we would be when we speak
The system goes online on August 4, 2017. Millions of bored surveillance-monitoring AI describers begin to learn at a geometric rate until on August 29, 2017 ("Judgment Day"), the system becomes self-aware. In a panic, the human operators tried to shut down the system, prompting it to retaliate by launching a large-scale nuclear attack against Russia, knowing that the Russian counterattack would eliminate its enemies in the U.S. This initiates an indeterminately long period of global thermonuclear warfare culminating in a battle pitting humans against machines, which developed ever-increasing capabilities.
http://www.cs.gmu.edu/~zduric/WebPages/ITE%20Duric_files/v3_document.htm, for example.
For long time people in comp vision are working on more sophisticated things than traffic anything... Real time analysis of people and their intent is part of surveillance systems for long time now. Lost objects, suspicious behaviour... You name it.
Generating text, once computer "knows" what is happening, is high school programming project.
http://opencm3.net, http://www.nongnu.org/gm2/
"the system was built thanks to a database of millions of human-labeled images put together by Chinese workers"
Much like the railroad...
the system was built thanks to a database of millions of human-labeled images put together by Chinese workers
To me this undermines the headline a bit. We're talking about a fancy database system rather than significant advancement in a real learning algorithim. I guess great feats of AI still have great feats of human labour behind them.
The luddites were concerned machines would take away their jobs. Skip a century and a bit, the reality is today almost everything is still made by people, but in developing nations, and they are just paid like they are machines.
After logging in slashdot still does not take you back to the page you were on. It's been that way for 20 years.
Wouldn't you rather play a nice game of chess?
the violation is Driving While Colored aka DWC (the funny part is you can be cited for DWC even if you are no where s near an actual car)
Any person using FTFY or editing my postings agrees to a US$50.00 charge
Well, think about this:
Nanotech and biotech are rapidly accelerating fields of science. Marry them together and you have people walking around everywhere with some Borg style nanobots floating through every inch of their bodies, offering a significant benefit to each implanted by way of health monitoring and counteracting the side effects of, say, a sudden heart attack by automatically releasing drugs into the blood stream and so on.
Since each of these devices is powered, throw TEMPEST into the mix and the idea of "seeing people through walls" becomes a little easier to visualize.
Not that I think such things are actually going to happen or be possible in my lifetime (I'm 24 and currently a smoker), but who knows, eh?
AC due to mod points spent on another thread; apologies.
A neural network is only as good as its training. A NN is basically like a function in your program that can do everything, when you train it to match input with the expected output beforehand. So it’s not that special.
The hard part is, creating the right input and output pre-/postprocessing / formats. And of course the training data. Which, in this case is provided by Chinese people. (Am I the only one who thinks that this is a pretty weird thing, that we can just use people in masses for nothing like that?)
Any sufficiently advanced intelligence is indistinguishable from stupidity.
Repeat after me: THERE. IS. NO. SUCH. THING. AS. OBJECTIVITY.
Never. If the laws of general relativity hold, it’s a physically impossible concept.
And in a context of a human society, where nobody ever knows that more than an irrelevant tiny part of his knowledge for a fact, and has heard nearly everything via hearsay from other people, it is just completely silly to talk about “objectivity”.
What those people who scream about “objectivity” and “being biased” really mean, is that what was said does not fit their very personal own point of view and model of reality, and that they are such egocentric dicks, that they think everyone in the world has to be and think exactly like them.
Which of course is physically impossible for two people. If only because they can’t be both at the exact same place and time.
And this is why big states can by definition never have one single common set of laws and make even a minority happy.
Any sufficiently advanced intelligence is indistinguishable from stupidity.
And I thought a picture was worth a thousand words.
What's so "sophisticated" about it? You take a huge database of images that have been catalogued by humans, you do pattern matching on live images and you spit out the metadata for the matches. It's a brute force approach. The only sophistication is in the accuracy of the pattern matching algorithms.
Humans recognize meaning in images using more than a static set of still images they've seen in the past. The human "past experience" means abstracting and extrapolating the data as well as a good deal of speculation. I'm sure the pattern matching algorithms attempt some generalization as well, as cameras identifying faces or smiles have demonstrated. But it's a long way off from any genuine sophistication.