AI Can Generate a 3D Model of a Person After Watching a Few Seconds of Video (sciencemag.org) 43
An anonymous reader shares a report: A new algorithm creates 3D models using standard video footage from one angle. The system has three stages. First, it analyzes a video a few seconds long of someone moving -- preferably turning 360-degree to show all sides -- and for each frame creates a silhouette separating the person from the background. Based on machine learning techniques -- in which computers learn a task from many examples -- it roughly estimates the 3D body shape and location of joints. In the second stage, it "unposes" the virtual human created from each frame, making them all stand with arms out in a T shape, and combines information about the T-posed people into one, more accurate model. Finally, in the third stage, it applies color and texture to the model based on recorded hair, clothing, and skin.
Re: (Score:2)
Every armchair genius in here commenting about how this is nothing new or impressive is completely missing that this creates a fully posable and articulated model with joints, etc.
Please show me a prior example of an existing solution that accomplishes the same thing from regular video footage.
I can see this being popular with the gamer crowd (Score:2)
Motion is still lacking (Score:2)
Based on the video, what we should really be working on is how to make such models move like a real human.
Re: (Score:1)
Based on the video, what we should really be working on is how to make such models move like a real human.
And yet another reason I grow to hate AI. It easily brings as much bad as good, if not more.
Simply put, once you have an accurate model of a human, that you can't easily tell from a real one, then you just green screen an actor with the same basic body shape and positioning pads in a camera, and we have "proof" of whatever you wanted proof of.
Perhaps you could still spot such fakes today, but for how much longer? And, really, how many people would notice even today? I know this isn't the first story that
Re: (Score:3)
Hating the inevitable doesn't seem very useful. It isn't "what if", it's "when".
Re: Motion is still lacking (Score:2)
Yes, hardware on blockchain, distributed so no one can be "ministry of truth". Video editing software then must publish their source videos, with veracity to be trusted, and final product also checksumed on chain.
Re: (Score:1)
Well, the terrorists in Douma didn't even need automatic 3D scan or AI. They only needed to spread the rumor of a chemical attack, film panicked people dousing each other with water, fetch a few corpses of babies and add them to the "movie stage".
This requires to have a few suffocating people (they had inhaled smoke due to being in a war zone) and a few dead babies around.
Then, everyone can call the US, UK and France for their bullshit. But will the journalist and their so-called free press dare question th
Re: (Score:2)
There exists some pretty compelling research on having a 3d model walk a complicated terrain in a natural looking fashion.
But I was unable to find the video :(. It probably isn't older than a year or so..
You call that AI? (Score:1)
Doesn't seem like it has much to do with artificial intelligence.
Neat scanner program, though.
Re: (Score:2)
Generalizing The Entire Person (Score:2)
...In the next stage the system uses the phenotype information shown in the video combined with the 99.9999% shared genetic information between all humans to create a nearly exact genome for the person. ...In the final step the AI system uses your likes, preferences, cookie trails and other information gleaned from your online life to fill in the life experience portion and recreate your personality.
This all allows the marketers to then pinpoint your exact weaknesses to advertising so that you BUY, BUY, BUY
Re: (Score:2)
Re: (Score:2)
Of course this is AI. (Score:1)
Watching a video and from that being able to build a 3d model of a person.... ....is something that normally requires intelligence. Rocks can't do it. Neither can plants. Nor steam engines.
This fits the definition of Artificial Intelligence exactly.
You just have some kind of chip on your shoulder.
AI porn (Score:2)
This is one step closer to the computer generated porn.
Re: (Score:2)
It isn't useful till it can differentiate the booty between Jessica Alba in Idle Hands and Jessica Biel. Until then body modeling is only a vague suggestion that people are different.
Locating the joints automatically is nust a step above marking them manually and using a generic body.
Expect Rapid Physical Modeling (Score:2)
This is actually one of the advances I had predicted many years ago. The next steps are interesting but ultimately you end up with what I refer to as "Rapid Physical Modeling" where in you can take an object, do a quick 360 view of it and then capture it's physical properties by manipulating it. A simple example is demonstrating how a potted plant can bend and it's ability to bend and spring back are inferred. The obvious benefit of this is that you can quickly model things with complex interactive respo
Why are these described as AI? (Score:2)
Even the summary describes it as a [learning] algorithm. Yes, it's bloody impressive, but it's an algorithm tuned to execute one specific task. If anything it's an Artificial Savant [wikipedia.org].
Contradiction (Score:2)
A new algorithm creates 3D models using standard video footage from one angle. The system has three stages. First, it analyzes a video a few seconds long of someone moving -- preferably turning 360-degree to show all sides
That's not one angle, then, is it?