Paul Debevec animates a photo-real digital face

86,234 views ・ 2009-10-20

TED


Please double-click on the English subtitles below to play the video.

00:18
One of the biggest challenges in computer graphics
0
18330
3000
00:21
has been being able to create a photo-real,
1
21330
2000
00:23
digital human face.
2
23330
2000
00:25
And one of the reasons it is so difficult is that, unlike aliens and dinosaurs,
3
25330
3000
00:28
we look at human faces every day.
4
28330
3000
00:31
They are very important to how we communicate with each other.
5
31330
2000
00:33
As a result, we're tuned in to the subtlest things
6
33330
3000
00:36
that could possibly be wrong with a computer rendering,
7
36330
2000
00:38
in order to believe whether these things are realistic.
8
38330
3000
00:41
And what I'm going to do in the next five minutes
9
41330
2000
00:43
is take you through a process
10
43330
2000
00:45
where we tried to create a reasonably photo-realistic computer-generated face,
11
45330
3000
00:48
using some computer graphics technology we've developed,
12
48330
3000
00:51
and also some collaborators at a company called Image Metrics.
13
51330
2000
00:53
And we're going to try to do a photo-real face
14
53330
2000
00:55
of an actress named Emily O'Brien, who is right there.
15
55330
3000
00:58
And that's actually a completely computer-generated rendering of her face.
16
58330
3000
01:01
By the end of the talk, we're going to see it move.
17
61330
2000
01:03
The way that we did this is we tried to start with Emily herself,
18
63330
3000
01:06
who was gracious enough to come to our laboratory
19
66330
3000
01:09
in Marina Del Rey, and sit for a session in Light Stage 5.
20
69330
3000
01:12
This is a face-scanning sphere, with 156 white LEDs all around
21
72330
4000
01:16
that allow us to photograph her
22
76330
2000
01:18
in a series of very controlled illumination conditions.
23
78330
3000
01:21
And the lighting that we use these days looks something like this.
24
81330
3000
01:24
We shoot all of these photographs in about three seconds.
25
84330
2000
01:26
And we basically capture enough information
26
86330
3000
01:29
with video projector patterns that drape over the contours of her face,
27
89330
4000
01:33
and different principle directions of light from the light stage,
28
93330
2000
01:35
to figure out both the coarse-scale
29
95330
2000
01:37
and the fine-scale detail of her face.
30
97330
2000
01:39
If we zoom in on this photograph right here,
31
99330
2000
01:41
we can see it's a really nice photograph to have of her,
32
101330
2000
01:43
because she is lit from absolutely everywhere at the same time
33
103330
4000
01:47
to get a nice image of her facial texture.
34
107330
2000
01:49
And in addition, we've actually used polarizers on all the lights --
35
109330
3000
01:52
just like polarized sunglasses can block
36
112330
2000
01:54
the glare off of the road,
37
114330
2000
01:56
polarizers can block the shine off of the skin,
38
116330
2000
01:58
so we don't get all those specular reflections to take this map.
39
118330
3000
02:01
Now, if we turn the polarizers around just a little bit,
40
121330
2000
02:03
we can actually bring that specular reflection
41
123330
2000
02:05
of the skin back in,
42
125330
2000
02:07
and you can see she looks kind of shiny and oily at this point.
43
127330
3000
02:10
If you take the difference between these two images here,
44
130330
3000
02:13
you can get an image lit from the entire sphere of light
45
133330
2000
02:15
of just the shine off of Emily's skin.
46
135330
3000
02:18
I don't think any photograph like this had ever been taken
47
138330
2000
02:20
before we had done this.
48
140330
2000
02:22
And this is very important light to capture,
49
142330
3000
02:25
because this is the light that reflects off the first surface of the skin.
50
145330
2000
02:27
It doesn't get underneath the translucent
51
147330
2000
02:29
layers of the skin and blur out.
52
149330
2000
02:31
And, as a result, it's a very good cue
53
151330
2000
02:33
to the detailed shape of the skin-pore structure
54
153330
3000
02:36
and all of the fine wrinkles that all of us have,
55
156330
2000
02:38
the things that actually make us look like real humans.
56
158330
2000
02:40
So, if we use information that comes off of this specular reflection,
57
160330
4000
02:44
we can go from a traditional face scan
58
164330
2000
02:46
that might have the gross contours of the face and the basic shape,
59
166330
4000
02:50
and augment it with information
60
170330
2000
02:52
that puts in all of that skin pore structure and fine wrinkles.
61
172330
3000
02:55
And, even more importantly,
62
175330
2000
02:57
since this is a photometric process that only takes three seconds to capture,
63
177330
3000
03:00
we can shoot Emily
64
180330
2000
03:02
in just part of an afternoon,
65
182330
2000
03:04
in many different facial poses and facial expressions.
66
184330
3000
03:07
So, here you can see her moving her eyes around, moving her mouth around.
67
187330
3000
03:10
And these we're actually going to use to create a photo-real digital character.
68
190330
3000
03:13
If you take a look at these scans that we have of Emily,
69
193330
4000
03:17
you can see that the human face does an enormous amount of amazing things
70
197330
3000
03:20
as it goes into different facial expressions.
71
200330
3000
03:23
You can see things. Not only the face shape changes,
72
203330
3000
03:26
but all sorts of different skin buckling and skin wrinkling occurs.
73
206330
4000
03:30
You can see that the skin pore structure changes enormously
74
210330
4000
03:34
from stretched skin pores
75
214330
2000
03:36
to the regular skin texture.
76
216330
2000
03:38
You can see the furrows in the brow and how the microstructure changes there.
77
218330
3000
03:41
You can see muscles pulling down at flesh to bring her eyebrows down.
78
221330
4000
03:45
Her muscles bulging in her forehead when she winces like that.
79
225330
3000
03:48
In addition to this kind of high-resolution geometry,
80
228330
3000
03:51
since it's all captured with cameras, we've got a great texture map to use for the face.
81
231330
4000
03:55
And by looking at how the different color channels of the illumination,
82
235330
2000
03:57
the red and the green and the blue,
83
237330
2000
03:59
diffuse the light differently,
84
239330
2000
04:01
we can come up with a way of shading the skin on the computer.
85
241330
3000
04:04
Then, instead of looking like a plaster mannequin,
86
244330
3000
04:07
it actually looks like it's made out of living human flesh.
87
247330
3000
04:10
And this is what we used
88
250330
2000
04:12
to give to the company Image Metrics
89
252330
2000
04:14
to create a rigged, digital version of Emily.
90
254330
3000
04:17
We're just seeing the coarse-scale geometry here.
91
257330
2000
04:19
But they basically created a digital puppet of her,
92
259330
3000
04:22
where you can pull on these various strings,
93
262330
2000
04:24
and it actually moves her face in ways that are
94
264330
2000
04:26
completely consistent with the scans that we took.
95
266330
2000
04:28
And, in addition to the coarse-scale geometry,
96
268330
2000
04:30
they also used all of that detail
97
270330
2000
04:32
to create a set of what are called "displacement maps"
98
272330
3000
04:35
that animate as well.
99
275330
3000
04:38
These are the displacement maps here.
100
278330
2000
04:40
And you can see those different wrinkles actually show up as she animates.
101
280330
3000
04:43
So the next process was then to animate her.
102
283330
2000
04:45
We actually used one of her own performances to provide the source data.
103
285330
4000
04:49
So, by analyzing this video with computer vision techniques,
104
289330
3000
04:52
they were able to drive the facial rig
105
292330
2000
04:54
with the computer-generated performance.
106
294330
2000
04:56
So what you're going to see now, after this,
107
296330
2000
04:58
is a completely photo-real digital face.
108
298330
2000
05:00
We can turn the volume up a little bit if that's available.
109
300330
2000
05:02
Emily: Image Metrics is a markerless, performance-driven animation company.
110
302330
4000
05:06
We specialize in high-quality facial animation
111
306330
3000
05:09
for video games and films.
112
309330
2000
05:11
Image Metrics is a markerless, performance-driven animation company.
113
311330
4000
05:15
We specialize in high quality facial animation
114
315330
3000
05:18
for video games and films.
115
318330
3000
05:21
Paul Debevec: So, if we break that down into layers, here's that diffuse component
116
321330
2000
05:23
we saw in the first slide.
117
323330
2000
05:25
Here is the specular component animating.
118
325330
2000
05:27
You can see all the wrinkles happening there.
119
327330
2000
05:29
And there is the underlying wireframe mesh.
120
329330
5000
05:34
And that is Emily herself.
121
334330
2000
05:36
Now, where are we going with this here?
122
336330
2000
05:38
We've gone a little bit beyond Light Stage 5. This is Light Stage 6,
123
338330
4000
05:42
and we're looking at taking this technology
124
342330
2000
05:44
and applying it to whole human bodies.
125
344330
2000
05:46
This is Bruce Lawmen, one of our researchers in the group,
126
346330
2000
05:48
who graciously agreed to get captured running in the Light Stage.
127
348330
3000
05:51
And let's take a look at a computer-generated version
128
351330
2000
05:53
of Bruce, running in a new environment.
129
353330
2000
06:01
And thank you very much.
130
361330
2000
06:03
(Applause)
131
363330
1000
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7