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

DeepMind的AI能根据你的描述直接合成图片 – 译学馆
未登录,请登录后再发表信息
最新评论 (0)
播放视频

DeepMind的AI能根据你的描述直接合成图片

DeepMind's AI Creates Images From Your Sentences | Two Minute Papers #163

亲爱的学霸们大家好 这里是由Károly Zsolnai-Fehér带来的两分钟论文
Dear Fellow Scholars, this is Two Minute Papers with Károly Zsolnai-Fehér.
这是一篇新发表的文章 出自于谷歌DeepMind公司一群想法疯狂的家伙们
This is one of those new, absolutely insane papers from the Google DeepMind guys.
你将会看到的 是针对一连串图像的算法的一些后续工作
You are going to see a followup work to an algorithm that looks at a bunch of images
它将自动学习鸟类 人脸 珊瑚礁等等概念
and from that, it automatically learns the concept of birds, human faces or coral reefs,
这样 我们就可以写一段新的文字说明 然后它可以
so much so that we’re able to write a new sentence, and it will generate a new, close
将文字说明转变成逼真的图像
to photorealistic image from this written description.
这个网络能够创造出和训练它时所用的图像
This network is capable of creating images that are significantly different than the
完全不同的图像
ones it has been trained on.
这听起来完全像是虚构出来的
This already sounds like science fiction.
和科幻小说一样
Completely unreal.
此项工作利用了PixelCNN技术
This work goes by the name PixelCNN.
我们过一会将讨论其后续的工作
We’ll discuss a followup work to that in a moment.
此算法的缺点在于 图像的像素是逐个生成的
The downside of this method is that these images are generated pixel by pixel, and many
同时 这些像素之间相互受到制约
of these pixels depend on their neighborhoods.
例如 当我开始绘制鸟喙的一处像素时
For instance, if I start to draw one pixel of the beak of a bird, the neighboring pixels
相邻的像素就会受到制约 必须画出鸟喙相邻的部分
have to adhere to this constraint and have to be the continuation of the beak.
很显然 这个图像包含很多小结构
Clearly, these images have a lot of structure.
这意味着我们不能将此过程并行化 每次只能为图片创建
This means that we cannot do this process in parallel, but create these new images one
一个像素
pixel at a time.
这是一个非常缓慢且计算量巨大的过程 因此
This is an extremely slow and computationally expensive process, and hence, the original
最初的论文里最多只能显示32×32和64×64格式的图像
paper showed results with 32×32 and 64×64 images at most.
当我们一步步处理运算过程时 会发现算法的执行时间与我们可以生成的像素数之间
As we process everything sequentially, the execution time of the algorithm scales linearly
是呈线性关系的
with the number of pixels we can generate.
就像一个拥有大量装配线的工厂 却只有一人
It is like a factory where there are a ton of assembly lines, but only one person to
在辛苦经营
run around and operate all of them.
我们的目标是 使这些关联不大的像素
Here, the goal was to start generating different regions of these images independently, but
能够独立生成不同区域的图像
only in cases when these pixels are not strongly correlated.
例如 不能使用相邻的像素
For instance, doing this with neighbors is a no-go.
这有可能会实现 但是也具有很大的挑战性 本文包含了 有关像素如何选择的细节
This is possible, but extremely challenging, and the paper contains details on how to select
以及判断它们何时是独立的具体说明
these pixels and when we can pretend them to be independent.
现在 你可以看到这些惊人的效果
And now, feast your eyes upon these spectacular results.
如果我们要寻找一只拥有黑色头 橙色眼睛 橙色喙的黄色的鸟
If we’re looking for “A yellow bird with a black head, orange eyes and an orange bill”,
我们就能看到更详细的图像
we’re going to see much more detailed images.
新算法的复杂程度随着像素数量呈对数相关 而不是线性相关
The complexity of the new algorithm scales with the number of pixels not linearly, but
这基本上 相当于已经赢得技术并行化方面的头奖
in a logarithmic manner, which is basically the equivalent of winning the jackpot in terms
这也导致速度加快了100多倍
of parallelization, and it often results in a more than 100 times speedup.
这里成为了不再是一个人 而是由多人大规模运行的工厂
This is a factory that’s not run by one guy, but one that works properly.
第一作者Scott Reed 在twitter上发表了更多惊人的成果
The lead author, Scott Reed has also published some more amazing results on twitter as well.
在这些成果中 我们可以看到通过网络生成图像的一系列演变
In these examples, we can see the evolution of the final image that is generated by the network.
作为未来成果中的一部分 这种感觉很酷
It is an amazing feeling to be a part of the future.
需要注意的是 这个想法还不完备 这只是我们计划的第一步
And note that there is a ton of challenges with the idea, this is one of those typical
执行才是最重要的
cases when the idea is only the first step, and execution is king.
请仔细阅读文章
Make sure to have a look at the paper for more details.
根据计划时间表 我们尽量每周推出两集
According to our regular schedule, we try our best to put out two videos every week.
每月八集
That’s eight episodes a month.
如果您喜欢这一集的内容 认为它值得花费一美元 请考虑在Patreon上
If you feel that eight of these episodes is worth a dollar for you, please consider supporting
支持我们
us on Patreon.
这样 我们将会为您提供更多的精彩内容!
This way, we can create more elaborate episodes for you.
感谢您的大力支持 使得我们的收视率不断攀升
The channel is growing at a remarkable rate, and your support has been absolutely amazing.
我们很荣幸能够受到您的青睐
I am honored to have an audience like you Fellow Scholars.
让我们离下个目标又近了一步
We are quite close to hitting our next milestone.
这个目标将会回馈给科学界
And this milestone will be about giving back more to the scientific community and empowering
并给予其他研究项目更多的帮助
other research projects.
我已经将Patreon的页面链接及详细说明放到视频描述栏
I’ve put a link to our Patreon page with the details in the video description.
感谢您的观看和支持 我们下期再见!
Thanks for watching and for your generous support, and I’ll see you next time!

发表评论

译制信息
视频概述

AI能够可以根据你的描述合成图像哟!搜索更加方便!

听录译者

收集自网络

翻译译者

[B]无牙无耳

审核员

豆子

视频来源

https://www.youtube.com/watch?v=9bcbh2hC7Hw

相关推荐