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

从单一深度图像完成语义场景补全 – 译学馆
未登陆,请登陆后再发表信息
最新评论 (0)
播放视频

从单一深度图像完成语义场景补全

Semantic Scene Completion From One Depth Image | Two Minute Papers

亲爱的学霸们 这是由Karoly Zsolnai-Feher带来的两分钟论文
Dear Fellow Scholars, this is Two Minute Papers with Károly Zsolnai-Fehér.
本期介绍的是深度神经网络的一项很棒的应用
This piece of work is an amazing application of deep neural networks, that performs semantic
从单一深度图像即可完成场景补全
scene completion from only one depth image.
这张彩图就是深度图 不同颜色表示
This depth image is the colorful image that you see here, where the colors denote how
物体距离观察者的远近
far away different objects are from our camera.
用消费级硬件就能创造出深度图 如微软的
We can create these images inexpensively with commodity hardware, for instance, Microsoft’s
Kinect 就有深度传感器 可用于制图
Kinect has a depth sensor that is suitable for this task.
场景补全的意思是可以从这张大量信息缺失的深度图中
The scene completion part means that from this highly incomplete depth information,
还原出房间的完整景象
the algorithm reconstructs the geometry for the entirety of the room.
甚至连带图中缺失和封闭起来的部分!
Even parts, that are completely missing from our images or things that are occluded!
计算机图形学研究者把该算法的输出称为一种立体表示
The output is what computer graphics researchers like to call a volumetric representation or
也就是体素的集合 本质上可看成许多乐高积木
a voxel array, which is essentially a large collection of tiny Lego pieces that build
堆成了一个场景
up the scene.
但还不全如此 因为语义场景补全意味着算法
But this is not all because the semantic part means that the algorithm actually understands
能理解场景 能给场景的不同部分分类
what we’re looking at, and thus, is able to classify different parts of the scene.
类别包括墙壁 窗口 地板 沙发 以及其他家具
These classes include walls, windows, floors, sofas, and other furniture.
场景补全和分类并非新技术 但该算法
Previous works were able to do scene completion and geometry classification, but the coolest
最酷的部分在于不仅改进了这些步骤 而且
part of this algorithm is that it not only does these steps way better, but it does them
能同时对它们进行计算
both at the very same time.
研究中使用了三维卷积神经网络
This work uses a 3D convolutional neural network to accomplish this task.
之所以要用三维 是要让算法能够处理立体数据
The 3D part is required for this learning algorithm to be able to operate on this kind of volumetric data.
如你所见 效果一级棒 和真实场景十分相近
As you can see, the results are excellent, and are remarkably close to the ground truth data.
不久前 第一次看到基于神经网络的从二维图像
If you remember, not so long ago, I flipped out when I’ve seen the first neural network-based
理解三维场景的技术时 我兴奋不已
techniques that understood 3D geometry from 2D images.
那项技术使用了生成对抗网络 要复杂得多
That technique used a much more complicated architecture, a generative adversarial network,
也没有做场景补全 而且输出的分辨率
which also didn’t do scene completion and on top of that, the resolution of the output
也低很多 直观上看就是说堆积的乐高积木块很大
was way lower, which intuitively means that Lego pieces were much larger.
真不可思议
This is insanity.
机器学习的研究进展惊人 即使是
The rate of progress in machine learning research is just stunning, probably even for you seasoned
看我们节目的各位高要求学霸也会震惊的
Fellow Scholars who watch Two Minute Papers and have high expectations.
前面几期我们介绍过多种关于神经网络的工作
We’ve had plenty of previous episodes about the inner workings of different kinds of neural networks.
我把链接放在视频介绍里 想提高机器学习水平的
I’ve put some links to them in the video description, make sure to have a look if you wish to brush
朋友们一定要看看
up on your machine learning kung fu a bit.
论文作者还发布了新的数据集 供将来研究使用
The authors also published a new dataset to solve these kind of problems in future research
而且该技术的输出可以与真实场景相比较
works, and, it is also super useful because the output of their technique can be compared to ground truth data.
将来有新的算法时 这个数据集可以作为比较的标杆
When new solutions pop up in the future, this dataset can be used as a yardstick to compare results with.
项目的源代码也开放了
The source code for this project is also available.
学霸们欢呼吧
Tinkerers rejoice!
感谢您的观看和支持 下次见喽
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

论文地址:http://sscnet.cs.princeton.edu/

听录译者

收集自网络

翻译译者

HX

审核员

知易行难

视频来源

https://www.youtube.com/watch?v=8YWgar0uCF8

相关推荐