High&NewTech:人工智能技术滥用之DeepNude技术(从下载致系统宕机→最后被禁用)而引发的AI道德底线的深度拷问—191017再次更新
相关文章 相关报道 DeepNude was a deepfakes application which allowed to alter photos of a person to make them appear nude, designed to work on females. Several days after its June 23rd, 2019, release and rapid climb to popularity, the app was permanently shut down over the concerns of potential misuse, although already downloaded copies of the software were continued to be shared. DeepNude是一款类似deepfake应用程序,它允许修改一个人的照片,使其看起来是裸体的,专为女性设计。该应用于2019年6月23日发布,并迅速走红。几天后,出于对潜在滥用的担忧,该应用程序被永久关闭。 Motherboard联系了DeepNude的创建者,此人化名为“阿尔贝托”(Alberto,匿名)。他表示,该软件基于加州大学伯克利分校研究者开发的开源算法pix2pix创建,并使用1万张女性LUO图加以训练。这一算法与之前的人工智能换脸技术deepfake算法相似,也和无人车技术所使用的算法相似。 DeepNude使用pix2pix开源算法分析输入图像,识别服装并删除它,用生成的裸体图像替换它。该算法在10000多张女性裸照的数据集上进行训练。 “阿尔贝托”表示,该软件目前之所以只能用于女性照片,是因为女性LUO体图像更容易在网上找到,但他希望能创建一个男性版本的软件。他还表示,他继续这一实验是出于“有意思”和好奇心。“我不是偷窥狂。我是技术的爱好者,”他说,“继续提升这个算法,最近也是吸取了之前的教训(来自其它创业公司)和经济上的问题,我问我自己是否可以在这个算法上获得经济回报。这就是我创建DeepNude的原因。” 但Motherboard指出,之前就有deepfake在网络上流传,但这种技术很容易就会成为伤害女性的工具——要么在未经同意的前提下使用女性照片,要么在网上恶意散布SQ内容。而DeepNude相当于deepfake技术的进化版,操作更简单,处理速度更快。所以,带来的危害更大。 其实,这款软件并非完美无缺。大多数图像,尤其是低分辨率图像,产生了一些视觉伪影。Deepnude完全失败了,因为有些照片使用了奇怪的角度、光线或衣服,这些照片似乎与它使用的神经网络脱节了。当我们给它一张卡通人物杰西卡·兔子的图像时,它会完全扭曲并破坏了图像。 网友评论:AI的换脸技术、脱衣技术,这件事,在今天可归纳为三句话:火是肯定要火的。乱是一定要乱的。如何监管,大概是不知道的。哦,对了,最后应该说一下如何防止别人做出你的AI换脸、脱衣视频?不要发太多自拍,即使自拍的时候,切记多穿点衣服。 基于AI算法的一键脱衣官宣关闭6月28日消息 此前,美国科技媒体Motherboard报道,Twitter用户@deepnudeapp 最近开发出一款名叫DeepNude的APP,只要给DeepNude输入一张女性照片,借助神经网络技术,App可以自动“脱掉”女性身上的衣服,显示出裸体。不过在6月28日凌晨,@deepnudeapp 宣布关闭这款App及网站。 Hi! DeepNude is offline. Why? Because we did not expect these visits and our servers need reinforcement. We are a small team. We need to fix some bugs and catch our breath. We are working to make DeepNude stable and working. We will be back online soon in a few days. ——06.27 最新的一条消息: Here is the brief history, and the end of DeepNude. We created this project for user's entertainment a few months ago. We thought we were selling a few sales every month in a controlled manner. Honestly, the app is not that great, it only works with particular photos. We never thought it would become viral and we would not be able to control the traffic. We greatly underestimated the request. Despite the safety measures adopted (watermarks) if 500,000 people use it, the probability that people will misuse it is too high. We don't want to make money this way. Surely some copies of DeepNude will be shared on the web, but we don't want to be the ones who sell it. Downloading the software from other sources or sharing it by any other means would be against the terms of our website. From now on, DeepNude will not release other versions and does not grant anyone its use. Not even the licenses to activate the Premium version. DeepNude软件相关评论 I’m glad DeepNude is dead. As a person and as a father, I thought this was one of the most disgusting applications of AI. To the AI Community: You have superpowers, and what you build matters. Please use your powers on worthy projects that move the world forward. --Andrew Ng
This is an “invasion of sexual privacy,” Danielle Citron, professor of law at the University of Maryland Carey School of Law, who recently testified to Congress about the deepfake threat, told Motherboard. Several media reports have noted how the app could be used to take a photo of a clothed woman and transform that into a nude image. Cyber Civil Rights Initiative (CCRI), which seeks protection against “revenge” porn tweeted, “This is a horrifically destructive invention and we hope to see you soon suffer consequences for your actions.” CCRI President Mary Anne Franks later tweeted, “It's good that it's been shut down, but this reasoning makes no sense. The app's INTENDED USE was to indulge the predatory and grotesque sexual fantasies of pathetic men.” DeepNude底层技术的其他正确应用DeepNude 软件主要使用了 Image Inpainting for Irregular Holes Using Partial Convolutions 中提出的Image-to-Image技术,该技术有很多其它的应用,比如把黑白的简笔画转换成色彩丰富的彩图,你可以点击下方的Image-to-Image Demo在浏览器中尝试Image-to-Image技术。 Deep Computer Vision in DeepNude1. Image Inpainting 图像修复 在 Image_Inpainting(NVIDIA_2018).mp4 视频中左侧的操作界面,只需用工具将图像中不需要的内容简单涂抹掉,哪怕形状很不规则,NVIDIA的模型能够将图像“复原”,用非常逼真的画面填补被涂抹的空白。可谓是一键P图,而且“毫无ps痕迹”。 该研究来自Nvidia的Guilin Liu等人的团队,他们发布了一种可以编辑图像或重建已损坏图像的深度学习方法,即使图像穿了个洞或丢失了像素。这是目前2018 state-of-the-art的方法。 2. Pix2Pix (need for paired train data) Image-to-Image Translation with Conditional Adversarial Networks 是伯克利大学研究提出的使用条件对抗网络作为图像到图像转换问题的通用解决方案。 3. CycleGAN (without the need for paired train data) CycleGAN使用循环一致性损失函数来实现训练,而无需配对数据。 换句话说,它可以从一个域转换到另一个域,而无需在源域和目标域之间进行一对一映射。 这开启了执行许多有趣任务的可能性,例如照片增强,图像着色,样式传输等。您只需要源和目标数据集。 Future1. Obj-GAN 事实上,我们不需要图像到图像。我们可以使用gan直接从随机值生成图像或从文本生成图像。\ 2. StoryGAN 进阶版神笔:只需一句话、一个故事,即可生成画面。 现在用得最多的Image-to-Image技术应该就是美颜APP了,所以可以去开发一个更加智能的美颜相机。 参考文章 |
|