ADM-201 dump PMP dumps pdf SSCP exam materials CBAP exam sample questions

真人的虚假视频–以及如何识别出来 – 译学馆
未登录,请登录后再发表信息
最新评论 (0)
播放视频

真人的虚假视频--以及如何识别出来

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

看看这些图片
Look at these images.
现在 告诉我哪个奥巴马是真的
Now, tell me which Obama here is real.
巴拉克•奥巴马:“为帮助家庭维持生计
(Video) Barack Obama: To help familiesrefinance their homes,
可以投资高科技制造业
to invest in thingslike high-tech manufacturing,
清洁能源 以及能创造好的工作岗位的基础设施行业”
clean energy and the infrastructurethat creates good new jobs.
Supasorn Suwajanakorn : 有吗
Supasorn Suwajanakorn: Anyone?
答案是 哪个也不是
The answer is none of them.
没有一个是真的
None of these is actually real.
我来告诉你 我们是怎样做到的
So let me tell you how we got here.
对这项工作 我的灵感来自于一个
My inspiration for this work was a project
对保留从大屠杀幸存者那里学习的最后机会很重要的项目
meant to preserve our last chance for learning about the Holocaust from the survivors.
项目被称作“证词的新维度”
It’s called New Dimensions in Testimony,
它能让你与一个真实的大屠杀幸存者
and it allows you to haveinteractive conversations
的全息影像互相交流
with a hologramof a real Holocaust survivor.
男人:你是怎么在大屠杀中活下来的
(Video) Man: How did yousurvive the Holocaust?
全息图像:我怎么活下来的?
(Video) Hologram: How did I survive?
我活下来 我相信 因为上天的庇护
I survived, I believe, because providence watched over me.
事实上 这些答案都是事先在录音室录制好的
Turns out these answerswere prerecorded in a studio.
没想到效果令人震惊
Yet the effect is astounding.
你感觉作为一个人 和他的故事以及他本人如此接近
You feel so connected to his story and to him as a person.
我认为关于人际互动 存在某些特殊的东西
I think there’s something specialabout human interaction
比书本 讲座 电影能教我们的
that makes it much more profound
更有深意且更个性化
and personal than what books or lectures or movies could ever teach us.
我看到这一点并开始思考
So I saw this and began to wonder,
我们能为任何人创造一个这样的模型吗
can we create a modellike this for anyone?
一个外表和言行举止像他们的模型
A model that looks, talksand acts just like them?
于是我开始思考这是否可行
So I set out to see if this could be done
最终想出了一个新的解决方案
and eventually came up with a new solution
可以只用下面这些制作一个人的模型:
that can build a model of a person using nothing but these:
某人现有的照片和视频
existing photos and videos of a person.
如果你能利用这类被动信息
If you can leveragethis kind of passive information,
只是那儿的照片和视频
just photos and video that are out there,
那些是推而广之的关键
that’s the key to scaling to anyone.
顺便一提 这是理查德•费曼
By the way, here’s Richard Feynman,
除了是诺贝尔物理学奖获得者
who in addition to being a Nobel Prize winner in physics
也被称作传奇的老师
was also known as a legendary teacher.
如果我们能让他回到课堂之上
Wouldn’t it be greatif we could bring him back
作演讲并激励上百万的孩子们不是很棒吗
to give his lecturesand inspire millions of kids,
也许不仅用英语 还可以用任一种语言
perhaps not just in Englishbut in any language?
抑或你可以向祖父母征求
Or if you could ask our grandparents
建议并听一些安慰的话
for advice and hear those comforting words
即使他们已不在人世
even if they’re no longer with us
或者用这个工具 书的作者不管活着与否
Or maybe using this tool,book authors, alive or not,
可以为任何感兴趣的读者大声朗读他们的作品
could read aloud all of their booksfor anyone interested.
其中创造性的可能性是无穷的
The creative possibilitieshere are endless,
对我来说 这很令人兴奋
and to me, that’s very exciting.
这就是它目前所能做的
And here’s how it’s working so far.
首先 我们来介绍一种
first we introduce a new technique that
能根据任何图像 甚至无需本人3D扫描的图像
can reconstruct a high-detailed 3D face model from any image
重构高清3D面部模型的新技术
without ever 3D-scanning the person.
这是来自不同视图的同样的输出模型
And here’s the same output modelfrom different views.
这项技术也适用于视频
This also works on videos,
通过在视频每一帧上运行相同的算法
by running the same algorithm on each video frame
生成一个活动的3D 模型
and generating a moving 3D model.
这是不同角度的输出模型
And here’s the sameoutput model from different angles.
事实上这个问题很有挑战性
It turns out this problemis very challenging,
但关键的技术是我们要分析
but the key trickis that we are going to analyze
预先准备好的此人的图片集
a large photo collectionof the person beforehand.
对小布什 我们只要在 Google上搜索一下就行
For George W. Bush,we can just search on Google,
由此 我们可以建立一个常规的模型
and from that, we are ableto build an average model,
一个可重复的 细化的模型在细微之处恢复表情
an iterative, refined modelto recover the expression
比如折痕和皱纹
in fine details,like creases and wrinkles.
在这点有趣的是
What’s fascinating about this is
这些照片集可以来自你独特的照片
that the photo collection can come from your typical photos.
你在做什么表情或在哪拍的照片
It doesn’t really matterwhat expression you’re making
并不重要
or where you took those photos.
重要的是 要具备很多这类要素
What matters isthat there are a lot of them.
而且我们还缺少色彩 所以接下来
And we are still missing color here, so next,
我们开发了一种新的混合技术
we develop a new blending technique
改进了单一的平均的方法
that improves upona single averaging method
并使面容纹理和颜色明朗起来
and produces sharpfacial textures and colors.
并且该技术可用于任何表情
And this can be done for any expression.
现在我们能进行个人的模型的控制
Now we have a controlof a model of a person,
该技术的控制方式基于静态照片的序列
and the way it’s controlled now is by a sequence of static photos.
注意皱纹出现并消失的方式 取决于表情的变化
Notice how the wrinkles come and go,depending on the expression.
我们还用视频来操纵模型
We can also use a videoto drive the model.
丹尼尔•克雷格:“对 但不知何故
(Video) Daniel Craig: Right, but somehow,
我们已经设法吸引了一些更了不起的人”
we’ve managed to attractsome more amazing people.
这儿还有个有趣的演示
SS:And here’s another fun demo.
你在这儿看到的都是可操纵的人物模型
So what you see hereare controllable models of people
基于互联网照片而制作
IP from their internet photos
现在 如果你从输入视频中传输动作
Now,if you transferthe motion from the input video,
我们真的可以操纵整个政党
we can actually drive the entire party.
乔治 w•布什:“这是一项难以通过的法案
George W. Bush:It’s a difficult bill to pass,
因为有很多不确定因素
because there’s a lot of moving parts,
立法过程可能是丑陋的”
and the legislative processes can be ugly.
回过头看看
So coming back a little bit,
我们的终极目标是抓住他们这些人
our ultimate goal, rather,is to capture their mannerisms
说话和微笑的特殊习惯及独特方式
or the unique way eachof these people talks and smiles.
为了做到这一点 我们真的可以
So to do that, can weactually teach the computer
通过一个人的录像片段
to imitate the way someone talks
来教会机器人模仿别人说话吗
by only showing itvideo footage of the person?
我所做的正是如此 我让电脑看了
And what I did exactly was,I let a computer watch
14小时的纯正的巴拉克•奥巴马演讲视频
14 hours of pure Barack Obamagiving addresses.
这是我们仅凭他的音频能做出来的视频
And here’s what we can producegiven only his audio.
奥巴马:“结果一目了然
(Video) BO: The results are clear.
美国商业在过去连续75月内创造了1450万
America’s businesses have created14.5 million new jobs
新的工作岗位”
over 75 straight months
这里合成的只是嘴部区域
So what’s being synthesized hereis only the mouth region,
我们是这样做的
and here’s how we do it.
我们利用神经网络来转换并输入音频
Our pipeline uses a neural network to convert and input audio
到嘴巴上的点
into these mouth points.
BO:“我们通过我们的工作 或通过医疗保险或医疗补助做到这点”
(Video) BO: We get it through our jobor through Medicare or Medicaid.
然后我们合成(面部)纹理 强化细节和牙齿
SS:Then we synthesize the texture,enhance details and teeth,
并将其与原视频中的头部和背景混合在一起
and blend it into the head and background from a source video.
BO:“女性可以得到免费的体检
(Video) BO: Women can get free checkups,
而且你不能因为是女性而多交钱
and you can’t get charged morejust for being a woman.
年轻人可以听取父母的安排 直到26岁”
Young people can stay on a parent’s plan until they turn 26.
我认为这些结果看起来非常逼真和有趣
SS:I think these resultsseem very realistic and intriguing,
但同时有些恐怖 甚至对我也是如此
but at the same time frightening even to me
我们的目标是建立个人的准确的模型
Our goal was to build an accurate model of a person,
而不是歪曲他们
not to misrepresent them.
但有一件事让我担心 它可能被滥用
But one thing that concerns meis its potential for misuse.
人们自Photoshop面世以来的很长一段时间
people been thinking about this problem for a long time
一直在思考这个问题
since the days when Photoshopfirst hit the market.
作为一名研究员 我也在研究对策技术
As a researcher, I’m also workingon countermeasure technology,
我是人工智能基金会持续努力的一员
and I’m part of an ongoingeffort at AI Foundation,
该基金会用机器学习和人类操控相结合
which uses a combinationof machine learning and human moderators
来学习和检测假图像和视频
to detect fake images and videos,
与我自身的工作进行对抗
fighting against my own work.
我们计划发布的其中一项工具称作 “现实守护者”
And one of the tools we plan to release is called Reality Defender,
它是一个网页浏览器插件 在浏览网页时
which is a web-browser plug-inthat can flag potentially fake content automatically,
自动标记潜在的假内容
right in the browser.
尽管如此 假视频甚至会在有人有机会核实之前
Despite all this, though, fake videos could do a lot of damage
就造成很大的损害
even before anyone has a chance to verify,
所以 让每个人都知道
so it’s very importantthat we make everyone aware
目前的可能性是很重要的
of what’s currently possible
我们可以怀有恰当的假设并对看到的东西注意鉴定
so we can have the right assumption and be critical about what we see.
在我们能充分模拟个体之前
There’s still a long way to go
在我们能确保这项技术的安全性之前
before we can fully model individual people
还有很长的路要走
and before we can ensurethe safety of this technology.
但我很兴奋也充满希望
But I’m excited and hopeful,
因为如果我们正确并小心地使用它
because if we use it right and carefully,
这个工具可以让世界上任何个体的积极影响
this tool can allow any individual’spositive impact on the world
得以大规模扩展
to be massively scaled
并真正帮助我们按照我们希望的方式塑造未来
and really help shape our future the way we want it to be.
谢谢大家
Thank you.

发表评论

译制信息
视频概述

通过研究员对真人的虚假视频的研究,推动假视频的仿真化,但同时也带来了一些隐患。

听录译者

收集自网络

翻译译者

全幼儿园最困

审核员

审核员 V

视频来源

https://www.youtube.com/watch?v=o2DDU4g0PRo

相关推荐