深度学习将灰度图着色_通过深度学习为视频着色

最新推荐文章于 2023-11-02 10:30:46 发布

weixin_26752765

最新推荐文章于 2023-11-02 10:30:46 发布

阅读量842

点赞数

文章标签：深度学习机器学习人工智能 tensorflow python

原文链接：https://towardsdatascience.com/colorize-a-video-with-deep-learning-15b30da3b57c

版权

深度学习将灰度图着色

零本地设置/ DeOldify / Colab笔记本 (Zero Local Setup / DeOldify / Colab Notebook)

Following up on my goal of doing things easily, I wanted to try DeOldify to convert B&W video content with audio over and over again, without having to change code or lose progress due to the free resource session timing out on Google Colab.

遵循我轻松做事的目标，我想尝试DeOldify一次又一次地转换B＆W视频内容和音频，而不必更改代码或由于Google Colab上的免费资源会话超时而失去进度。

Below are the steps — around 10 clicks, enabling you to do just that! Keep in mind I have tested only short videos (5 mins) with the free hardware resources.

以下是步骤-大约10次点击，使您能够做到这一点！请记住，我仅使用免费的硬件资源测试了短片(5分钟)。

Open the Colab Notebook link hosted in the repository. This article might also familiarize you with the power of Colab for your next exciting AI paper.

打开存储库中托管的Colab Notebook链接。本文还可能使您熟悉Colab在下一篇令人兴奋的AI论文中的功能。

https://colab.research.google.com/github/ojjsaw/video-processing/blob/master/Custom_DeOldify_VideoColorizer_Colab.ipynb

https://colab.research.google.com/github/ojjsaw/video-processing/blob/master/Custom_DeOldify_VideoColorizer_Colab.ipynb

First, ensure a free GPU resource is allocated by logging in. Click on Runtime menu on the top > Change runtime type > GPU > Save and then click Connect on the top right.

首先，确保通过登录分配免费的GPU资源。单击顶部的“运行 时”菜单> 更改运行时类型 > GPU > 保存，然后单击右上角的“ 连接 ”。

After logging in, hit Connect again and update the URL to your video link obtained from the share icon on your YouTube video. The default URL is the iconic “Arrival of the Train” clip from Lumière Brothers’ 1895 — History of Films 101.

登录后，再次单击“ 连接” ，然后将URL更新为从YouTube视频上的共享图标获得的视频链接。默认的URL是LumièreBrothers的1895年-电影史101中的标志性“火车到达”剪辑。

You can optionally modify the default google drive export directory path.

您可以选择修改默认的Google驱动器导出目录路径。

Hit the Run icon on the left! Every time you open this page (session), you will be asked to authorize google drive to mount it to your current session.

点击左侧的“运行”图标 ！每次您打开此页面( 会话 )时，系统都会要求您授权google驱动器将其安装到当前会话中。

Copy-paste the authorization code and hit Enter. Click on the Setup text section to bring it in focus and hit the Run After option from the Runtime menu on the top of the page.

复制并粘贴授权码，然后按Enter 。单击“ 设置”延伸部分以使其突出显示，然后单击页面顶部“ 运行时” 菜单中的“运行后”选项。

Sit tight and go grab a coffee! Just make sure your machine is running and the browser stays alive.

坐好去喝杯咖啡！只要确保您的机器正在运行并且浏览器保持运行状态即可。

It will take a while to process, but after a successful conversion, the source and result videos will be uploaded to your google drive at the chosen path.

处理需要一些时间 ，但是成功转换后，源视频和结果视频将按照所选路径上传到您的Google驱动器。

Again, above is the simplest path to conversion without the bells and whistles. For more customization, please take a look at the original notebook.

同样，以上是最简单的转换方法，没有花哨的时间。要进行更多定制，请查看原始笔记本。

In future posts, we will try to increase the frames-per-second of a video for a smoother clip and possibly even upscale the video to a higher resolution to satisfy the requirements of current generation display devices.

在以后的文章中，我们将尝试增加视频的每秒帧数以实现更流畅的剪辑，甚至可能将视频放大到更高的分辨率，以满足当前显示设备的要求。

If you’d like to try more start-to-finish quick AI experiments, visit my last post: Train a Neural Network to classify images & optimize CPU inferencing in 10 mins

如果您想尝试更多的从头到尾的快速AI实验，请访问我的上一篇文章： 训练神经网络对图像进行分类并在10分钟内优化CPU推理