Click on any phrase to play the video from that point.
[man] So why don't we move again into the graphics realm.
Adobe has a long history of working in the realm of graphics;
Photoshop, of course, a ground-breaking program.
We're going to see a little sneak on about searching through your photos
and finding commonality and finding photos that relate to one another--
something that we all sort of deal with as we take thousands and thousands of photos
with our digital cameras.
[Rainn Wilson] Yeah, I have that problem, because I literally have like--
seriously, I have like 2,000-3,000 photos on my laptop.
[man] Absolutely.
[Rainn] And sometimes it's overwhelming, like I look, I'm like, "Oh, geez."
[man] That's right.
[Rainn] I can't look through these.
[man] So we're going to have Jon Brandt come out and show us something
called Pixel Nuggets for how you can start to find commonality in your photos.
Jon?
[Rainn] Pixel Nuggets? [man] Pixel Nuggets.
[♪blues music♪]
[Rainn] I saw them in concert.
[♪blues music♪]
[Jon Brandt] Hello, everybody.
I'm Jon Brandt, and I'm here to show you some really cool visual search technology
that's been developed in the research lab along with a colleague of mine, Zhe Lin,
and some folks in Hamburg, Germany led by (s/l Hart Metharnk).
So like everybody else, I have thousands of photos on my laptop as well,
and I have a few indexed here just to show you the random selection of pictures
that I chose for this demo,
and I'm going to show you how we can use visual search to help organize our photos.
So visual search just means using images to find other images
based on the visual qualities in the image.
So as an example, I'm going to pick this sunflower,
and I just clicked on it and the interface changes so that
the search query that I clicked on shows up over here on the left
and all of the relevant pictures show on the right,
a kind of ranked list that we're all used to looking at,
and the results actually look pretty good.
We've got these nice, pretty flowers centered in the frame,
but it's not perfect.
There are, for instance, these buttercups interfere,
and those are clearly not belonging in the sunflower realm.
And the reason that happened is because we just told it
we are interested in this image as a whole, and we would like--
we're really interested in the sunflower, so I'm going to give it a hint
by drawing a rectangle around the sunflower
and it'll go off and do a localized object search instead.
[Rainn] Wow! Come on!
[applause]
[Jon] Thank you.
Now, what we have is all sunflowers, you'll notice,
and you also, if you look carefully, you'll see that the sunflowers
are appearing in different contexts, the lighting's different, color's different,
and also these little green boxes--I should not point my finger, but actually, the mouse--
around the found object.
So it works well for flowers, but that's not all.
Let me show you something else.
So it also is very effective at landmarks,
and you guys have probably seen something similar to this
if you've used Google Image Search.
If you put the Golden Gate Bridge in, you'll get lots of pictures
and one thing to notice about this result
is that they're all kind of similar; they're all basically the same viewpoint,
the same lighting and so forth,
and perhaps as a designer, I want to find other types of images of the Golden Gate Bridge
because we can use that same trick by drawing a rectangle
this time around the stanchion of the bridge,
and now, what I've got is lots of different views of the Golden Gate Bridge.
In fact, this black and white view from very oblique angle
and this one where-.
[Rainn] Who's she? Who's that?
[Jon] Yeah, I don't know.
I wish I'd got her name. [laughter]
Now I'm off track!
Let me go back to my demo. [laughter]
[Rainn] You look a little like Toby.
[Jon] Okay, landmarks--people!
[Rainn] You're doing great. This is incredible.
I can't believe this, seriously.
[Jon] So we can also use this to find people.
For instance, I clicked on that girl as a query,
and you can see that there are lots of photos in this particular collection
of these family members all kind of posed around the same area,
and this is actually pretty effective for grouping together photos that have been taken
at the same session or in the same setting.
And one thing to notice, also, is that the girl is ranked first
even though there are other very similar images,
and like, if I click on the sky, the same thing will happen.
[Rainn] What happens if you click on the boobs of the woman?
[laughter]
[Jon] Not in this demo! [laughter]
Later.
[Rainn] Yeah.
[Jon] I haven't had any beer yet, so--.
Okay, so people--let's see, where's--I've been side-tracked so much,
I don't know how much time I've got, but anyway-.
[Rainn] Take your time, dude.
[Jon] Okay. [Rainn] We've got all night.
[Jon] Cats.
So it likes cats, and you'll notice that similar to before,
we get a lot of related background, but let's say we're really just interested in pictures
where the cat's in a particular pose,
so I'm going to draw a rectangle around its face.
And you'll see that now, the kitty is looking straight at the camera,
and--well, mostly--and also it's appearing in different content.
Okay, I think one more thing.
This Is the product placement portion of my talk.
Let's say I had a video or I was interested in finding placements of logos and so forth.
I clicked on this image of the Coke can, and you can see that there's other stuff there,
and it did a pretty good job of finding Coke cans, but for instance, this image here
with the fence or the the pilars came up
because of the other aspects of the image.
So I'm going to narrow that by just drawing a rectangle around the Coke can logo,
and you can see that now, it's again localized, like I showed you before,
and there's actually this really interesting image here
which has both a Fanta and a Coke, and like before, if I draw a rectangle around the Coke,
I'll find the Coke can, including this image we had in the path.
But if I do the same thing with Fanta
just to show that it's real, it's finding the other object using the same image
as the original query, just [inaudible].
[Rainn>] Cool. Wow. [applause]
[Jon] So that's my speech. [applause]
Thank you.
[man] Thanks.
That was great, thank you!
So imagine [inaudible], if that sort of technology
being applied to your home image library.
It really gives you a whole new way to think about sorting and looking at your pictures,
and it's that sort of advanced research that we really do at Adobe
to try to really bring about new ways for people to work with media.
[Rainn] Cool!
