All language subtitles for 09-Lecture 1 Segment 9 What can AI do Perception.en

Okay, other areas. Another big sub-area of
artificial intelligence is perceiving the world,

and in large part this is vision
but there's other kinds of perception.

So, things like object recognition, face
recognition.

You probably have a lot of this
technology closer than you think, you

probably have face recognition built into your
cameras, that's actually face detection,

not always face recognition though some cameras do
that too. Segmenting scenes into pieces,

figuring out for a given image what it means,
what's going on.

Here's an example on the left of an
image, and overlaid on the image is the

machine's reconstruction of kind of the
underlying 3D

outline and mesh,

and you can see that reconstructed on
the right. This is from an image, but actually

it turns out that in addition to being
able to do a bunch of cool things in

vision with the image,

one realization we've had in cases like autonomous driving and vision is

we don't have to use the tools that humans
use. We've spent a long time with vision

just trying to use like a camera,
or maybe two cameras slightly apart because

that's what we have, we've got two cameras
slightly apart. But then we realized,

we can do other stuff. So what's this, anybody recognize this?

It's a Kinect. The Kinect's got sensors that you don't.

Sorry you didn't get a rangefinder, a depth detector, you just didn't.

But, you know, we can build them so why
not. And so, now we can do cool things like

take an image and produce a depth map
that isn't just about parallax, looking at

the difference between the two eyes, or
about kind of inferring from occlusion.

You'll notice, people like to think
that vision is all about having two images,

but if you close one eye, you can still see depth.

It's not like the world suddenly goes flat and you shriek. I mean, you close one eye, you still have a sense of depth,

we want to be to build machines that do that,
but at the moment we do pretty well by

using things like depth detectors, cause why not.

Let's take a look at science fiction again.

Does anybody recognize this movie? Does anybody know what this is gonna be?

Yeah, this is Terminator here, and let's
take a look at what it's like to be a

Terminator--it's relevant to vision. So here's what it's like to be a terminator.

It's actually a lot like being Governor of
California, apparently.

Okay, so he looks around,

okay, motorcycle, motorcycle, motorcycle, car,

motorcycle,

some place, target acquired. Okay, so

looking around, outlines, detection.
Identifying what the objects are, figuring

out what the target is, that's in the
movies.

Straight out of science fiction.

Let's look at some vision
recognition system--this is a cute demo from Al Rahimi's lab.

So here we have a camera panning around, and it's kind of--we can do exactly the same

thing but for real. So here we have

the cat.

Cat...

Frog...

Fox...

Dalmatian...

Bulldog... Terminate bulldog, right. Okay.

So,

this is a case where I think it's amazing
how close we can to what people thought

it might be like if this technology
were possible. This is not robots from the

future detecting the bulldog, this is today.