一鍵「復活」蒙娜麗莎！三星公布AI演算法：一張圖片即可生成視頻

The AI that can make Mona Lisa smile: Samsung reveals algorithm that creates fake talking-head videos using just ONE photo

能讓蒙娜麗莎微笑的人工智慧：三星公布了一種演算法，僅用一張照片就可以創建假的會說話的視頻

來源：每日郵報 翻譯：世界播

As if the world of deep-faked pictures and video wasn"t scary enough, researchers from Samsung"s AI center in Moscow have demonstrated an algorithm that can fabricate videos using only one image.

就好像世界上的深度偽造圖片和視頻還不夠可怕一樣，來自莫斯科三星人工智慧中心的研究人員展示了一種只用一張圖片就能製作視頻的演算法。

In a video demonstration and a paper published in the pre-print journal ArXiv, the researchers show the capabilities of what is described as "one-shot" and "few-shot" machine learning.

在一段視頻演示和發表在預印本網站ArXiv上的一篇論文中，研究人員展示了所謂的單樣本和少樣本機器學習的能力。

The results of their system bring to life popular faces like those of surrealist painter Salvador Dali and actress Marilyn Monroe using a single still image.

他們的系統僅使用單一靜止圖像，結果就能讓像超現實主義畫家薩爾瓦多?達利和女演員瑪麗蓮?夢露等流行面孔變得生動起來。

The more images that are fed into the program, the more realistic the resulting video becomes.

輸入程序的圖像越多，生成的視頻就越逼真。

Though a single image translated into a moving face may look noticeably altered, a sample of 32 images produces a moving picture with near lifelike accuracy.

雖然將一張圖像轉換成一張動態的臉可能看起來會發生明顯的變化，但是一個32張圖像的樣本可以生成一幅幾乎逼真的動態圖像。

Researchers" technology appears to take algorithms used to produce deepfakes far beyond existing methods not just in terms of realism, but in the amount of work that goes into creating photorealistic fakes.

研究人員的技術似乎不僅在現實感方面，而且在製作逼真假貨的工作量方面，都遠遠超出了現有方法用於製作深度偽造的演算法。

Current AI systems are usually trained on an extensive catalog, requiring the algorithm to scan large data sets of a subjects body or face.

目前的人工智慧系統通常在一個廣泛的目錄上進行訓練，要求演算法掃描對象的身體或面部的大數據集。

This method limits the system"s application to individuals who have sufficient amounts of pictures and videos available for use.

這種方法將系統的應用限制在有足夠數量的圖片和視頻可供使用的個人身上。

In the case of the researchers new software, however, those data sets are a more of a luxury than a necessity.

然而，就研究人員開發的新軟體而言，這些數據集與其說是必需品，不如說是奢侈品。

The newest algorithm was trained using the publicly available VoxCeleb database which contains more than 7,000 images of celebrities from YouTube videos.

最新的演算法是使用公開的VoxCeleb資料庫進行訓練的，該資料庫包含了來自YouTube視頻的7000多張名人圖片。

Using that data set, the AI was able to establish what researchers call "landmark" features-- universally identifiable traits -- among subjects" noise, eyes, and more.

利用這些數據集，人工智慧能夠在對象的噪音、眼睛和其他方面建立研究人員所說的「標誌性」特徵——普遍可識別的特徵。

Because of the algorithm"s recognizes common characteristics as opposed to specific traits of a subject, it"s able to quickly extrapolate images with little input and in some cases, bring to life still images of people like Fyodor Dostoevsky, who died in 1881 and was conceivably never captured on video.

由於該演算法識別的是一個對象的普通特徵，而不是特定特徵，因此它能夠在很少輸入的情況下快速推斷出圖像，在某些情況下，還能使像費奧多爾·陀思妥耶夫斯基這樣的人的靜態圖像生動起來，費奧多爾·陀思妥耶夫斯基死於1881年，可想而知，世上肯定沒有他的動態視頻。

This method also means that the technology is applicable toward non-celebrities.

這種方法也意味著該技術適用於非名人。

Using just a few selfies from a subject, a demonstration shows how the algorithm can create a video facsimile mimicking how they might move their head and speak.

僅用受試者的幾張自拍照，演示了該演算法如何創建一個視頻仿製品，模仿他們可能移動頭部和說話方式的樣子。

The AI is currently only able to produce "talking head" style videos of subjects from the shoulders up, but if the algorithm is anything like similar deepfake systems, that could quickly change.

目前，人工智慧只能生成受試者肩膀以上的「發言者頭部特寫」風格的視頻，但如果該演算法與深度偽造系統類似，情況可能很快就會改變。

Another recent and startling technology recently demonstrated by former employees of Google and Microsoft shows how an algorithm is able to convincingly replicate the voice of popular podcaster, Joe Rogan.

最近，谷歌和微軟的前僱員展示了另一項令人震驚的新技術，該技術展示了一種演算法能夠逼真地複製流行播客主持人喬?羅根的聲音。

Skeptics of deepfake technology say those two technologies combined would allow bad actors to feasibly fake videos that could spread misinformation on politicians or steal other victims" identity.

對深度偽造技術持懷疑態度的人表示，這兩種技術的結合將使不法分子有可能偽造視頻，傳播有關政客的假信息，或竊取其他受害者的身份。

Deepfaking technology has already been used at length to superimpose the face of celebrities on the bodies of porn actors causing some to call for legislation that would ban the use.

已經有人利用深度偽裝技術將名人的臉疊加到色情演員的身體上，導致很多人呼籲立法禁止使用這種技術。

喜歡這篇文章嗎？立刻分享出去讓更多人知道吧！