ABOUT THE SPEAKER
Supasorn Suwajanakorn - Computer scientist
Supasorn Suwajanakorn works on ways to reconstruct, preserve and reanimate anyone -- just from their existing photos and videos.

Why you should listen

Can we create a digital avatar that looks, acts and talks just like our sweet grandma? This question has inspired Supasorn Suwajanakorn, a recent PhD graduate from the University of Washington, to spend years developing new tools to make it a reality. He has developed a set of algorithms that can build a moving 3D face model of anyone from just photos, which was awarded the Innovation of the Year in 2016. He then introduced the first system that can replicate a person's speech and produce a realistic CG-animation by only analyzing their existing video footage -- all without ever bringing in the person to a Hollywood capture studio.

Suwajanakorn is working in the field of machine learning and computer vision. His goal is to bring vision algorithms out of the lab and make them work in the wild.

More profile about the speaker
Supasorn Suwajanakorn | Speaker | TED.com
TED2018

Supasorn Suwajanakorn: Fake videos of real people -- and how to spot them

蘇帕索恩蘇瓦耶納柯恩: 真人的假影片以及如何辨視出它們

Filmed:
1,453,308 views

你認為你很擅長辨識出名人說出他們沒有說的話的那些假影片嗎?來看看這場驚人的演說和技術展示,了解怎麼辦到的。電腦科學家和研究生蘇帕索恩蘇瓦耶納柯恩說明他如何用人工智慧和 3D 建模,使影片與聲音同步,來創造出如相片般真實的假影片。他也進一步談到這項技術的道德意涵和可能創意,以及要採取什麼步驟,來避免它被誤用。
- Computer scientist
Supasorn Suwajanakorn works on ways to reconstruct, preserve and reanimate anyone -- just from their existing photos and videos. Full bio

Double-click the English transcript below to play the video.

00:12
Look at these images圖片.
0
876
1151
看看這些影像。
00:14
Now, tell me which哪一個 Obama奧巴馬 here is real真實.
1
2051
2635
告訴我,當中哪一個
歐巴馬是真的?
00:16
(Video視頻) Barack巴拉克 Obama奧巴馬: To help families家庭
refinance再融資 their homes家園,
2
4710
2861
(影片)歐巴馬:
協助家庭做房款再融資,
00:19
to invest投資 in things
like high-tech高科技 manufacturing製造業,
3
7595
2647
投資於高科技製造業之類、
00:22
clean清潔 energy能源
4
10266
1159
乾淨能源
00:23
and the infrastructure基礎設施
that creates創建 good new jobs工作.
5
11449
2779
及能夠創造新工作機會的基礎建設。
00:26
SupasornSupasorn SuwajanakornSuwajanakorn: Anyone任何人?
6
14647
1484
講者:有人知道嗎?
00:28
The answer回答 is none沒有 of them.
7
16155
1874
答案是:通通不是。
00:30
(Laughter笑聲)
8
18053
1114
(笑聲)
00:31
None沒有 of these is actually其實 real真實.
9
19191
1786
這些通通不是真的。
00:33
So let me tell you how we got here.
10
21001
1840
讓我告訴各位,
我們如何走到這一步。
00:35
My inspiration靈感 for this work
11
23940
1578
我做出這項作品的靈感
00:37
was a project項目 meant意味著 to preserve保留 our last
chance機會 for learning學習 about the Holocaust大屠殺
12
25542
5411
是一項計畫,計畫原來的目的是要
保存我們從大屠殺生還者那裡
了解大屠殺的最後機會。
00:42
from the survivors倖存者.
13
30977
1768
00:44
It's called New Dimensions尺寸 in Testimony證詞,
14
32769
2627
它叫做「證詞的新維度」,
00:47
and it allows允許 you to have
interactive互動 conversations對話
15
35420
3126
它讓你可以和大屠殺生還者的
00:50
with a hologram全息照相
of a real真實 Holocaust大屠殺 survivor倖存者.
16
38570
2556
立體投影進行互動式交談。
00:53
(Video視頻) Man: How did you
survive生存 the Holocaust大屠殺?
17
41793
1966
(影片)男子:你如何
從大屠殺活下來?
00:55
(Video視頻) Hologram全息照相: How did I survive生存?
18
43783
1668
(影片)投影:我如何活下來?
00:57
I survived倖存,
19
45912
1807
我能活下來,
01:00
I believe,
20
48419
1527
我相信,
01:01
because providence普羅維登斯 watched看著 over me.
21
49970
3023
是因為有老天保佑我。
01:05
SSSS: Turns out these answers答案
were prerecorded預錄 in a studio工作室.
22
53573
3454
講者:其實這些答案都是
在攝影棚預先錄影好的。
01:09
Yet然而 the effect影響 is astounding驚人.
23
57051
2452
但效果仍然很驚人。
01:11
You feel so connected連接的 to his story故事
and to him as a person.
24
59527
3619
你會覺得和他的故事
及他本人很有連結。
01:16
I think there's something special特別
about human人的 interaction相互作用
25
64011
3301
我認為,人類互動中有著某種特性,
01:19
that makes品牌 it much more profound深刻
26
67336
2757
它會讓使用者的體驗相當深切,
01:22
and personal個人
27
70117
2198
相當個人化,
01:24
than what books圖書 or lectures講座
or movies電影 could ever teach us.
28
72339
3485
而這些是書本、課程,或電影
沒有辦法教導我們的。
01:28
So I saw this and began開始 to wonder奇蹟,
29
76267
2425
我看到這個之後,開始納悶,
01:30
can we create創建 a model模型
like this for anyone任何人?
30
78716
2810
我們能不能為任何人
創造這樣的模型?
01:33
A model模型 that looks容貌, talks會談
and acts行為 just like them?
31
81550
2975
和他們有相似外表、
說話方式,和行為的模型?
01:37
So I set out to see if this could be doneDONE
32
85573
2007
所以,我打算試試能不能辦到,
01:39
and eventually終於 came來了 up with a new solution
33
87604
2310
最終,我找出了一個新的解決方案,
01:41
that can build建立 a model模型 of a person
using運用 nothing but these:
34
89938
3220
只要用下列這些東西,
就能建造出一個人的模型:
01:45
existing現有 photos相片 and videos視頻 of a person.
35
93747
2214
一個人既有的照片和影片。
01:48
If you can leverage槓桿作用
this kind of passive被動 information信息,
36
96701
2617
若你能發揮這類被動資訊的功效,
01:51
just photos相片 and video視頻 that are out there,
37
99342
2007
只要既有的照片和影片,
01:53
that's the key to scaling縮放 to anyone任何人.
38
101373
2056
那就是能夠將規模擴大
到所有人的關鍵。
01:56
By the way, here's這裡的 Richard理查德 Feynman費曼,
39
104119
1777
順道一提,這是理察費曼,
01:57
who in addition加成 to being存在
a Nobel諾貝爾 Prize winner優勝者 in physics物理
40
105920
3413
他不只是諾貝爾物理獎的得主,
02:01
was also known已知 as a legendary傳奇的 teacher老師.
41
109357
2453
也是一位傳奇性的老師。
02:05
Wouldn't豈不 it be great
if we could bring帶來 him back
42
113080
2198
如果我們能找他回來教課,
02:07
to give his lectures講座
and inspire啟發 millions百萬 of kids孩子,
43
115302
3265
鼓舞數百萬名孩子,不是很棒嗎?
02:10
perhaps也許 not just in English英語
but in any language語言?
44
118591
2992
且不只是用英文,
還可以用任何語言?
02:14
Or if you could ask our grandparents祖父母
for advice忠告 and hear those comforting欣慰的 words
45
122441
4602
或是,我們的祖父母已經不在了,
還能問問他們意見,
聽他們說些撫慰的話,如何?
02:19
even if they're no longer with us?
46
127067
1770
02:21
Or maybe using運用 this tool工具,
book authors作者, alive or not,
47
129683
3396
或許能用這樣的工具,
讓不論還活著或已故的作家
02:25
could read aloud高聲 all of their books圖書
for anyone任何人 interested有興趣.
48
133103
2937
出來朗讀他們的書
給想要聽的讀者聽。
02:29
The creative創作的 possibilities可能性
here are endless無窮,
49
137199
2437
這個工具有著無限的創意可能性,
02:31
and to me, that's very exciting扣人心弦.
50
139660
1713
我對此感到十分興奮。
02:34
And here's這裡的 how it's working加工 so far.
51
142595
2002
目前,它的運作方式如下:
02:36
First, we introduce介紹 a new technique技術
52
144621
1667
首先,我們先採用一項新技術,
02:38
that can reconstruct重建 a high-detailed高詳細
3D face面對 model模型 from any image圖片
53
146312
4572
它能從任何影像,重新建構出
一個人非常細節的 3D 面部模型,
02:42
without ever 3D-scanningD 掃描 the person.
54
150908
2119
不需要對他做實際的 3D 掃瞄。
02:45
And here's這裡的 the same相同 output產量 model模型
from different不同 views意見.
55
153890
2642
這是從不同視角,輸出的類似模型。
02:49
This also works作品 on videos視頻,
56
157969
1502
它也能用在影片上,
02:51
by running賽跑 the same相同 algorithm算法
on each video視頻 frame
57
159495
2852
對每一格影片進行同樣的演算法,
02:54
and generating發電 a moving移動 3D model模型.
58
162371
2222
產生出會動的 3D 模型。
02:57
And here's這裡的 the same相同
output產量 model模型 from different不同 angles.
59
165538
2772
這是從不同的角度,輸出的類似模型。
03:01
It turns out this problem問題
is very challenging具有挑戰性的,
60
169933
2534
結果發現,這個問題非常有挑戰性,
03:04
but the key trick
is that we are going to analyze分析
61
172491
2525
但關鍵的技巧在於,我們得要
03:07
a large photo照片 collection採集
of the person beforehand預先.
62
175040
2966
事先分析同一個人的大量照片集。
03:10
For George喬治 W. Bush襯套,
we can just search搜索 on Google谷歌,
63
178650
2539
如果要做小布希總統,
我們用 Google 搜尋就可以了,
03:14
and from that, we are able能夠
to build建立 an average平均 model模型,
64
182309
2499
這樣,我們就可以建立
一個平均的模型,
03:16
an iterative迭代, refined精製 model模型
to recover恢復 the expression表達
65
184832
3111
一個經過反覆運算、改良後的模型,
它能夠很精密地重現出表情,
03:19
in fine details細節,
like creases摺痕 and wrinkles皺紋.
66
187967
2336
如皺褶和皺紋等。
03:23
What's fascinating迷人 about this
67
191326
1403
很炫的一點是,
03:24
is that the photo照片 collection採集
can come from your typical典型 photos相片.
68
192753
3423
只要用你一般的照片集
就可以做到了。
03:28
It doesn't really matter
what expression表達 you're making製造
69
196200
2603
你在照片中做什麼表情都無所謂,
03:30
or where you took those photos相片.
70
198827
1885
在任何地方拍的照片都行。
03:32
What matters事項 is
that there are a lot of them.
71
200736
2400
重要的是,要有很多張照片。
03:35
And we are still missing失踪 color顏色 here,
72
203160
1736
這裡,我們還缺了顏色,
03:36
so next下一個, we develop發展
a new blending混紡 technique技術
73
204920
2348
所以,下一步,我們開發出
一種新的混和技術,
03:39
that improves提高 upon
a single averaging平均 method方法
74
207292
2836
它是從單一平均化法改良而來的,
03:42
and produces產生 sharp尖銳
facial面部 textures紋理 and colors顏色.
75
210152
2818
能夠產生出精細的
面部質感和顏色。
03:45
And this can be doneDONE for any expression表達.
76
213779
2771
任何表情都可以使用。
03:49
Now we have a control控制
of a model模型 of a person,
77
217485
2499
現在我們可以控制一個人的模型,
03:52
and the way it's controlled受控 now
is by a sequence序列 of static靜態的 photos相片.
78
220008
3795
控制它的方式,是透過
一系列靜態照片來進行。
03:55
Notice注意 how the wrinkles皺紋 come and go,
depending根據 on the expression表達.
79
223827
3126
注意看,根據不同表情,
皺紋有時會出現有時會消失。
04:00
We can also use a video視頻
to drive駕駛 the model模型.
80
228109
2746
我們也可以用影片來驅動模型。
04:02
(Video視頻) Daniel丹尼爾 Craig克雷格: Right, but somehow不知何故,
81
230879
2593
(影片)丹尼爾克雷格:
對,但不知何故
04:05
we've我們已經 managed管理 to attract吸引
some more amazing驚人 people.
82
233496
3771
我們更能吸引到很了不起的人。
04:10
SSSS: And here's這裡的 another另一個 fun開玩笑 demo演示.
83
238021
1642
講者:這是另一個有趣的展示。
04:11
So what you see here
are controllable可控制 models楷模
84
239687
2246
各位在這裡看到的是可控制的模型,
04:13
of people I built內置
from their internet互聯網 photos相片.
85
241957
2444
我用他們在網路上的照片建立的。
04:16
Now, if you transfer轉讓
the motion運動 from the input輸入 video視頻,
86
244425
2904
如果把輸入影像的動作做轉換,
04:19
we can actually其實 drive駕駛 the entire整個 party派對.
87
247353
2152
我們就可以讓所有人跟著動起來。
04:21
George喬治 W. Bush襯套:
It's a difficult bill法案 to pass通過,
88
249529
2172
喬治布希:要通過
這個法案很困難,
04:23
because there's a lot of moving移動 parts部分,
89
251725
2303
因為有很多會變動的部分,
04:26
and the legislative立法 processes流程 can be ugly醜陋.
90
254052
5231
且立法過程有時是很醜陋的。
04:31
(Applause掌聲)
91
259307
1630
(掌聲)
04:32
SSSS: So coming未來 back a little bit,
92
260961
1837
講者:先倒帶一下,
04:34
our ultimate最終 goal目標, rather,
is to capture捕獲 their mannerisms裝相
93
262822
3191
我們的終極目標,
其實是要捕捉他們的動作習性,
04:38
or the unique獨特 way each
of these people talks會談 and smiles笑容.
94
266037
3045
或是說每個人說話
和微笑的獨特方式。
04:41
So to do that, can we
actually其實 teach the computer電腦
95
269106
2313
為了做到這一點,我們要教導電腦
04:43
to imitate模擬 the way someone有人 talks會談
96
271443
2222
去模仿一個人說話的方式,
04:45
by only showing展示 it
video視頻 footage鏡頭 of the person?
97
273689
2420
做法是把那個人的影片給電腦看。
04:48
And what I did exactly究竟 was,
I let a computer電腦 watch
98
276898
2577
我所做的其實就是讓電腦觀看
04:51
14 hours小時 of pure Barack巴拉克 Obama奧巴馬
giving addresses地址.
99
279499
3277
十四小時的影片,
內容全是歐巴馬在演講。
04:55
And here's這裡的 what we can produce生產
given特定 only his audio音頻.
100
283443
3516
只要提供他的聲音,
我們就能產生出這樣的成果。
04:58
(Video視頻) BOBO: The results結果 are clear明確.
101
286983
1777
(影片)歐巴馬:結果很清楚。
05:00
America's美國 businesses企業 have created創建
14.5 million百萬 new jobs工作
102
288784
4349
在 75 個月內,
美國的企業已經創造出了
05:05
over 75 straight直行 months個月.
103
293157
2774
1450 萬個新工作機會。
05:07
SSSS: So what's being存在 synthesized綜合 here
is only the mouth region地區,
104
295955
2905
講者:這裡合成的只有嘴部區域,
05:10
and here's這裡的 how we do it.
105
298884
1540
我們的做法是這樣的。
05:12
Our pipeline管道 uses使用 a neural神經 network網絡
106
300764
1826
我們的指令傳遞途徑
使用的是類神經網路,
05:14
to convert兌換 and input輸入 audio音頻
into these mouth points.
107
302614
2936
將輸入的聲音轉換成那些嘴部的點。
05:18
(Video視頻) BOBO: We get it through通過 our job工作
or through通過 Medicare醫保 or Medicaid醫療補助.
108
306547
4225
(影片)歐巴馬:透過我們的運作
或醫療保險、醫療補助來達成。
05:22
SSSS: Then we synthesize合成 the texture質地,
enhance提高 details細節 and teeth,
109
310796
3420
講者:接著我們再合成肌理,
強化細節和牙齒,
05:26
and blend混合 it into the head
and background背景 from a source資源 video視頻.
110
314240
3074
將它結合到頭部
以及來源影片的背景上。
05:29
(Video視頻) BOBO: Women婦女 can get free自由 checkups體檢,
111
317338
1905
(影片)歐巴馬:
女性能接受免費檢查,
05:31
and you can't get charged帶電 more
just for being存在 a woman女人.
112
319267
2968
你不會因為身為女性,
就要付比較多錢。
05:34
Young年輕 people can stay
on a parent's父母 plan計劃 until直到 they turn 26.
113
322973
3306
年輕人在 26 歲前都可以
掛在一位家長的方案底下。
05:39
SSSS: I think these results結果
seem似乎 very realistic實際 and intriguing奇妙,
114
327267
2952
講者:我認為這些結果
非常真實且有意思,
05:42
but at the same相同 time
frightening可怕的, even to me.
115
330243
3173
但同時,連我也覺得它們很嚇人。
05:45
Our goal目標 was to build建立 an accurate準確 model模型
of a person, not to misrepresent歪曲 them.
116
333440
4015
我們的目標是要建造出很精確的
人類模型,而不是要故意歪曲他。
05:49
But one thing that concerns關注 me
is its potential潛在 for misuse濫用.
117
337956
3111
但我擔心的一件事情,
是這項工具可能被誤用。
05:53
People have been thinking思維
about this problem問題 for a long time,
118
341958
2971
從 Photoshop 剛出現
在市場上之後,
05:56
since以來 the days when PhotoshopPhotoshop中
first hit擊中 the market市場.
119
344953
2381
長久以來大家都一直
在想著這個問題。
05:59
As a researcher研究員, I'm also working加工
on countermeasure對策 technology技術,
120
347862
3801
身為研究者,
我也在研究反制的技術,
06:03
and I'm part部分 of an ongoing不斷的
effort功夫 at AIAI Foundation基礎,
121
351687
2942
我參與了人工智慧
基金會的一項計畫,
06:06
which哪一個 uses使用 a combination組合
of machine learning學習 and human人的 moderators版主
122
354653
3397
結合機器學習和人類調節
06:10
to detect檢測 fake images圖片 and videos視頻,
123
358074
2144
來偵測出仿冒的影像和影片,
06:12
fighting戰鬥 against反對 my own擁有 work.
124
360242
1514
及對抗我自己的作品。
06:14
And one of the tools工具 we plan計劃 to release發布
is called Reality現實 Defender辯護人,
125
362675
3190
我們打算要推出的工具之一,
叫做「真相守衛者」,
06:17
which哪一個 is a web-browser網路瀏覽器 plug-in插入
that can flag potentially可能 fake content內容
126
365889
4039
它是瀏覽器的插件,
能標出有可能是仿冒的內容,
06:21
automatically自動, right in the browser瀏覽器.
127
369952
2533
全自動化,就裝在瀏覽器上。
06:24
(Applause掌聲)
128
372509
4228
(掌聲)
06:28
Despite儘管 all this, though雖然,
129
376761
1453
不過,儘管有這些努力,
06:30
fake videos視頻 could do a lot of damage損傷,
130
378238
1840
仿冒影片仍能造成許多傷害,
06:32
even before anyone任何人 has a chance機會 to verify校驗,
131
380102
3294
甚至在任何人有機會
驗證之前,傷害就已造成,
06:35
so it's very important重要
that we make everyone大家 aware知道的
132
383420
2722
所以,非常重要的是,
要讓大家都能意識到,
06:38
of what's currently目前 possible可能
133
386166
2007
目前有什麼是可能的,
06:40
so we can have the right assumption假設
and be critical危急 about what we see.
134
388197
3369
我們因此得以有正確的假設,
並對所看到的內容抱持批評的態度。
06:44
There's still a long way to go before
we can fully充分 model模型 individual個人 people
135
392423
5007
要能為個人完整地建模
還有很長的一段路要走,
06:49
and before we can ensure確保
the safety安全 of this technology技術.
136
397454
2786
也還要很多努力才能確保
這項科技的安全性。
06:53
But I'm excited興奮 and hopeful有希望,
137
401097
1587
但我很興奮,也抱著希望,
06:54
because if we use it right and carefully小心,
138
402708
3539
因為如果我們能正確、
小心地使用它,
06:58
this tool工具 can allow允許 any individual's個人
positive impact碰撞 on the world世界
139
406271
4309
這工具能讓任何人正面地影響世界,
07:02
to be massively大規模 scaled縮放
140
410604
2190
大規模地影響,
07:04
and really help shape形狀 our future未來
the way we want it to be.
141
412818
2742
且能真正能用我們希望的方式
來型塑我們的未來。
07:07
Thank you.
142
415584
1151
謝謝。
07:08
(Applause掌聲)
143
416759
5090
(掌聲)
Translated by Lilian Chiu
Reviewed by Yi-Fan Yu

▲Back to top

ABOUT THE SPEAKER
Supasorn Suwajanakorn - Computer scientist
Supasorn Suwajanakorn works on ways to reconstruct, preserve and reanimate anyone -- just from their existing photos and videos.

Why you should listen

Can we create a digital avatar that looks, acts and talks just like our sweet grandma? This question has inspired Supasorn Suwajanakorn, a recent PhD graduate from the University of Washington, to spend years developing new tools to make it a reality. He has developed a set of algorithms that can build a moving 3D face model of anyone from just photos, which was awarded the Innovation of the Year in 2016. He then introduced the first system that can replicate a person's speech and produce a realistic CG-animation by only analyzing their existing video footage -- all without ever bringing in the person to a Hollywood capture studio.

Suwajanakorn is working in the field of machine learning and computer vision. His goal is to bring vision algorithms out of the lab and make them work in the wild.

More profile about the speaker
Supasorn Suwajanakorn | Speaker | TED.com

Data provided by TED.

This site was created in May 2015 and the last update was on January 12, 2020. It will no longer be updated.

We are currently creating a new site called "eng.lish.video" and would be grateful if you could access it.

If you have any questions or suggestions, please feel free to write comments in your language on the contact form.

Privacy Policy

Developer's Blog

Buy Me A Coffee