Adobe概述 (Adobe Overview)

The design process changes from person to person, practice to practice, and profession to profession. As we advance further into the 21st century, the design process for many people leans heavily on technology. Adobe leads the charge in developing creative software for those looking to design and create through digital means, through the Creative Cloud suite.

設計過程因人而異，因實踐而異，因專業而異。 隨著我們進一步進入21世紀，許多人的設計過程在很大程度上依賴于技術。 土坯 負責為希望通過數字方式通過Creative Cloud套件進行設計和創建的人員開發創意軟件。

Adobe creates a wide variety of tools that address the needs of many creative genres. But what do Adobe products have to do with data science or machine learning?

Adobe創建了各種各樣的工具來滿足許多創意類型的需求。 但是Adobe產品與數據科學或機器學習有什么關系？

Well, this blog will discuss how Adobe incorporates machine learning / AI into a few of its products, in order to make the design process faster and more intuitive. We will discuss features that harness AI powered by Adobe Sensei in these products below in the Adobe Creative Cloud suite:

好吧，這個博客將討論Adobe如何將機器學習/ AI集成到其一些產品中，以使設計過程更快，更直觀。 我們將在以下Adobe Creative Cloud套件中的以下產品中討論利用Adobe Sensei支持的AI的功能：

• Premiere Pro and After Effects are video editors and might help YouTubers and streaming service content creators (like Netflix) to combine, rearrange, and edit video content.

  Premiere Pro和After Effects是視頻編輯器，可以幫助YouTuber和流服務內容創建者(例如Netflix)組合，重新排列和編輯視頻內容。

• Audition is an audio editor and might help a podcaster edit their audio file to add transitioning sound effects, or slicing out certain portions of the podcast before sharing.

  Audition是音頻編輯器，可以幫助播客編輯其音頻文件以添加過渡音效，或在共享之前將播客的某些部分切出。

• Photoshop is a photo editor and an essential tool for professional photographers to make adjustments to their images, whether it’s brushing away wrinkles on someone’s face, increasing the brightness, or montaging other content into the image.

  的Photoshop 是照片編輯器，是專業攝影師調整其圖像的必不可少的工具，無論它是刷去人臉上的皺紋，增加亮度還是將其他內容蒙版到圖像中。

• Illustrator is a must-have for graphic designers and UI/UX designers and provides a platform to design vector graphics such as logos, branding, digital art, posters, or website buttons.

  Illustrator是圖形設計師和UI / UX設計師的必備工具，并且提供了一個平臺來設計矢量圖形，例如徽標，品牌，數字藝術，海報或網站按鈕。

• InDesign assists in the design, layout, and publication of multi-page documents, which might be used by people on marketing teams, presentation designers, or magazine/publication editors.

  InDesign協助設計，布局和發布多頁文檔，營銷團隊，演示文稿設計師或雜志/出版物編輯人員可能會使用它們。

Premiere Pro和After Effects (Premiere Pro & After Effects)

內容感知填充 (Content-Aware Fill)

In After Effects, we can get rid of unwanted objects in our video footage using Adobe Sensei AI. The Content-Aware Fill tool in After Effects simply asks us for the region and the duration for the software to “fill” the video frames to mask things we don’t want to see. The tool then samples surrounding contextual pixels to generate pixel patterns in the video frames that “blend in” with the scene — as if the object never existed. This AI is probably built using Generative Adversarial Networks (GANs) — the same deep learning algorithms that can create incredibly convincing deepfakes.

在After Effects中，我們可以使用Adobe Sensei AI消除視頻素材中不需要的對象。 After Effects中的“內容感知填充”工具僅要求我們提供軟件“填充”視頻幀以掩蓋我們不愿看到的東西的區域和持續時間。 然后，該工具對周圍的背景像素進行采樣，以在與場景“融合”的視頻幀中生成像素模式-好像該對象從不存在。 該AI可能是使用生成對抗網絡(GAN)構建的 -相同的深度學習算法，可以創建令人信服的Deepfake 。

As a (very) concise overview — a GAN is composed of two competing neural networks: a generator and a discriminator. The generator looks at “true” data (the pixels of the road) and seeks to deceive the discriminator with “fake” generated data (pixels to match the road to obscure the car), whereas the mission of the discriminator is to detect the “fake” data. So after many rounds of training these neural nets, ultimately the goal is to have the generator get better and better at fooling the discriminator so that the discriminator cannot tell the “true” data from the “fake” data. The generator tries to maximize classification error, the discriminator tries to minimize classification error.

作為(非常)簡潔的概述，GAN由兩個相互競爭的神經網絡組成： 生成器和鑒別器 。 生成器查看“真實”數據(道路像素)，并試圖用“假”生成的數據(與道路相匹配的像素來掩蓋汽車)來欺騙鑒別器，而鑒別器的任務是檢測“偽造”數據。 因此，在對這些神經網絡進行了多輪訓練之后，最終目標是使生成器在欺騙鑒別器方面變得越來越好，以使鑒別器無法從“偽”數據中分辨出“真實”數據。 生成器嘗試使分類誤差最大化，鑒別器嘗試使分類誤差最小化。

變形切割 (Morph Cut)

In Premiere Pro, Morph Cut allows us to more smoothly transition between video cuts — particularly when interviewing someone. We might want to take out a few seconds of video when the interviewee is scratching their nose or something, but we want our video interview to look like we never cut anything out. We don’t want it to jump from one cut to another harshly, nor do we want to visibly dissolve or fade into the next cut. We want to transition as seamlessly as possible.

在Premiere Pro中，Morph Cut允許我們在視頻剪輯之間進行更平穩的過渡，尤其是在采訪某人時。 當受訪者抓撓鼻子或其他東西時，我們可能希望拍攝幾秒鐘的視頻，但我們希望視頻采訪看起來像我們從來沒有剪出任何東西。 我們既不希望它從一個切口急劇跳到另一個切口，也不希望明顯地溶解或淡入下一個切口。 我們希望盡可能無縫地過渡。

So, a Morph Cut is placed between two video cuts. Then, a duration for the transition is set (maybe 15, 30, or 60 frames). Finally, Adobe Sensei AI built into Premiere Pro analyzes the scene — the interviewee’s face, the surroundings, the light, the color, etc…and it generates video content to “blend” between the two cuts. Like the Content-Aware Fill tool mentioned before, this AI tool is likely achieved through GANs. It is difficult to see the smooth transition that Morph Cut provides without a “before and after” example, so check out this YouTube video by Adobe Creative Cloud that really shows off the effect.

因此，將“變形剪切”放置在兩個視頻剪切之間。 然后，設置過渡的持續時間(可能是15、30或60幀)。 最終，Premiere Pro中內置的Adobe Sensei AI分析了場景-受訪者的臉，周圍環境，光線，顏色等……并生成視頻內容以在兩個剪輯之間“融合”。 像前面提到的Content-Aware Fill工具一樣，此AI工具很可能是通過GAN實現的。 如果沒有“之前和之后”的示例，很難看到Morph Cut提供的平滑過渡，因此請觀看 Adobe Creative Cloud的YouTube視頻 ， 該視頻真正展示了這種效果。

自動改組 (Auto Reframe)

In Premiere Pro, Auto Reframe automatically changes the proportions of the video frame without losing the “important” focus of the footage. So for example, if we filmed a ship cruising in the San Francisco Bay using a DSLR camera with a 16:9 ratio and wanted to reframe the footage for a vertically oriented iPhone, we could use Auto Reframe to convert the 16:9 ratio to 9:16 and have it track the ship in the cropped footage. Check out this YouTube video again by Adobe Creative Cloud that shows you how to use this tool.

在Premiere Pro中，自動重畫框會自動更改視頻幀的比例，而不會丟失素材的“重要”焦點。 因此，例如，如果我們使用比例為16：9的DSLR相機拍攝了在舊金山灣巡航的船只，并想對垂直方向的iPhone的鏡頭進行重新構圖，則可以使用自動重新構圖將16：9的比例轉換為9:16并在裁剪后的畫面中跟蹤船。 再次觀看 Adobe Creative Cloud的YouTube視頻 ，向您展示如何使用此工具。

試鏡 (Audition)

混音 (Remix)

The Remix function in Adobe Audition automatically fits a music clip to any desired length. This video by Mike Murphy very succinctly demos and explains the Audition Remix tool. Imagine that you want a 30-second intro jingle for your podcast show. You scraped the internet and found a royalty-free music clip that you like, but it’s only 10 seconds long. Once again, it’s Adobe Sensei AI to the rescue! Once Remix is enabled and has analyzed the sound clip, you can stretch it out to 30 seconds, and the trained AI is able to automatically stitch together pieces of the 10-second music clip, to seamlessly extend it to your desired 30-second duration. Done!

Adobe Audition中的“混音”功能會自動將音樂剪輯調整為任何所需的長度。 此視頻由Mike Murphy制作 非常簡潔地演示并解釋了Audition Remix工具。 想象一下，您需要30秒的播客節目簡介。 您刮擦了互聯網，找到了自己喜歡的免版稅音樂剪輯，但它只有10秒長。 再一次，是要拯救Adobe Sensei AI！ 一旦啟用了Remix并分析了聲音片段，您就可以將其延長到30秒，并且受過訓練的AI可以自動將10秒音樂片段的片段縫合在一起，以將其無縫擴展到所需的30秒持續時間。 做完了！

自動引導 (Autoducking)

Autoducking uses AI to automatically adjust the volume up or down on audio. This is typically done for audio that uses both music and dialogue. For example, we might want the music softer in the background to hear the dialogue, but when the dialogue stops, we can increase the music volume. In this example, Adobe Sensei analyzes the locations and durations of sound files in Audition to understand and predict where the music or dialogue audio should be quieter or louder for the optimal audio experience. Check out Mike Murphy’s YouTube demo of the Autoducking feature in Adobe Audition.

Autoducking使用AI來自動調節音頻的音量。 通常對同時使用音樂和對話的音頻進行此操作。 例如，我們可能希望背景音樂更柔和以聽見對話，但是當對話停止時，我們可以增加音樂音量。 在此示例中，Adobe Sensei分析了Audition中聲音文件的位置和持續時間，以了解和預測音樂或對話音頻在何處應該更安靜或更響，以獲得最佳的音頻體驗。 查看Mike Murphy的YouTube演示 ， 演示 Adobe Audition中的Autoducking功能。

的Photoshop (Photoshop)

內容感知填充 (Content-Aware Fill)

Just like Content-Aware Fill in After Effects, Photoshop uses Adobe Sensei to generate “fill” pixels for a selected area of an image with pixel data sampled from adjacent portions of the image being edited. Want a pimple gone on your face in your latest selfie? AI in the form of Content-Aware Fill is there to save the day! It’ll sample the skin around the pimple, and make it disappear.

就像內容感知的After Effects中的填充一樣，Photoshop使用Adobe Sensei生成圖像的選定區域的“填充”像素，并從正在編輯的圖像的相鄰部分中采樣像素數據。 想要在最新的自拍中將粉刺涂在臉上嗎？ 以內容感知填充的形式提供的AI可以節省一天！ 它會采樣丘疹周圍的皮膚，并使其消失。

面部識別液化 (Face-Aware Liquify)

This feature automatically detects faces in the image and then lets us use custom grips (one for the mouth, one for the eyes, etc…) to alter the facial expressions. Adobe Sensei likely uses neural networks (computer vision) to find the faces and the features on the face to manipulate. This feature is quickly demonstrated in the first 30 seconds of this video tutorial by the Photoshop Training Channel.

此功能會自動檢測圖像中的臉部，然后讓我們使用自定義握柄(一個用于嘴巴，一個用于眼睛，等等)來更改面部表情。 Adobe Sensei可能使用神經網絡(計算機視覺)來找到面部和面部特征以進行操作。 在本視頻教程的前30秒中可以快速演示此功能 由Photoshop培訓頻道提供。

對象選擇 (Object Selection)

Object Selection makes lassoing and selecting elements in an image faster. Let’s say we took a photo of our friend against a blank white wall in her apartment, and our friend wants to look like she’s living her best life at the beach instead. So using Photoshop, we have to change the background around her. Using the Object Selection tool, we can simply click and drag a loose selection rectangle around her, and Adobe Sensei’s AI magic automatically draws a very accurate and tight selection that closely follows the profile of her body, so we can easily drop in the beach background! Mike Murphy shows the power of this feature in his YouTube demo here.

對象選擇使套索和選擇圖像中的元素更快。 假設我們在她公寓的空白白墻上拍下了我們朋友的照片，而我們朋友想看起來像是她在海灘上過著最好的生活。 因此，使用Photoshop，我們必須更改她周圍的背景。 使用“對象選擇”工具，我們只需單擊并在她周圍拖動一個松散的選擇矩形，Adobe Sensei的AI魔術就會自動繪制一個非常準確且緊緊的選擇，使其緊貼她的身體輪廓，因此我們可以輕松地掉入海灘背景！ Mike Murphy在此處的YouTube演示中展示了此功能的強大功能。

插畫家 (Illustrator)

全局編輯 (Global Editing)

Global Editing powered by Adobe Sensei in Illustrator is able to automatically detect similar vector geometries and lets us make changes to all of these alike geometries in a single act. This lets us make changes to reoccurring shapes faster. For example, if we designed wallpaper with a repeating flower pattern, and this flower pattern was arrayed 100 times on the canvas, we could make a change to one of the flower patterns using Global Editing, and all 100 copies of this pattern would be updated! This is clearly illustrated in this YouTube video by Adobe Creative Cloud.

由Adobe Sensei in Illustrator提供支持的Global Editing能夠自動檢測相似的矢量幾何形狀，并使我們能夠在一次操作中對所有這些相似的幾何形狀進行更改。 這使我們能夠更快地更改重復出現的形狀。 例如，如果我們設計了具有重復花朵圖案的墻紙，并且該花朵圖案在畫布上排列了100次，則可以使用“全局編輯”更改其中一種花朵圖案，并且將更新該圖案的所有100個副本！ Adobe Creative Cloud在此YouTube視頻中清楚地說明了這一點 。

InDesign (InDesign)

查找相似字體 (Find Similar Fonts)

Within InDesign, Adobe Sensei can analyze certain traits of a font — like the font’s width, serif, or other attributes. Then, it is able to recommend other fonts on your computer that are similar to the selected font. This recommendation system makes it a lot easier to search for more typography options with a particular font style. Check out this video here for a demo.

在InDesign中，Adobe Sensei可以分析字體的某些特征-例如字體的寬度，襯線或其他屬性。 然后，可以在計算機上推薦與所選字體相似的其他字體。 該推薦系統使搜索具有特定字體樣式的更多字體選項變得容易得多。 在此處查看此視頻以獲取演示。

內容感知適合 (Content-Aware Fit)

A common struggle in InDesign is placing, cropping, and scaling multiple photos in a layout quickly. Adobe Sensei streamlines this process. Once the layout is set up with empty frames where you want the images to go, the images can be dragged and dropped from a folder on your computer into the empty boxes. With those frames selected, we can use the Content-Aware Fit tool to let Adobe Sensei find the “best” portion of each image and scale it within the frame automatically. And of course, if you want to make further adjustments, you can spend more time and tweak it afterward. See this demo on YouTube by Adobe Creative Cloud to see this tool in action.

InDesign中的一個常見難題是快速在布局中放置，裁剪和縮放多張照片。 Adobe Sensei簡化了此過程。 一旦在布局中設置了要放置圖像的空白框，就可以將圖像從計算機上的文件夾拖放到空白框中。 選擇這些框架后，我們可以使用Content-Aware Fit工具讓Adobe Sensei找到每個圖像的“最佳”部分，并在框架內自動縮放比例。 當然，如果您想進行進一步的調整，則可以花更多的時間進行調整。 在 Adobe Creative Cloud的YouTube上觀看此演示，以了解此工具的實際效果。

下一步是什么？ (What’s Next?)

Adobe remains dominant in building software for creative means, but to stay on top, what’s the next step? As artificial intelligence and machine learning become more prolific in software, Adobe should continue to invest in Adobe Sensei’s capabilities to streamline the digital creative process as much as possible, while being careful not to take away useful features.

Adobe在為創意手段開發軟件方面仍然占主導地位，但要保持領先地位，下一步是什么？ 隨著人工智能和機器學習在軟件中的應用越來越廣泛，Adobe應該繼續投資Adobe Sensei的功能，以盡可能地簡化數字創作過程，同時注意不要剝奪有用的功能。

AI has the potential to achieve a particular result in a faster way through classification, prediction, and automation. Aggregating feedback data from the customer base — the photographers, the graphic designers, the video editors, and others, will be crucial to understand what people want to be automated. For example, Adobe might take note in situations where a skilled Photoshop user might exclaim:

AI具有通過分類，預測和自動化以更快的方式實現特定結果的潛力。 匯總來自客戶群(攝影師，圖形設計師，視頻編輯者等)的反饋數據對于了解人們想要自動化的內容至關重要。 例如，在熟練的Photoshop用戶可能驚呼的情況下，Adobe可能會注意：

“Gah, it be SO nice if Photoshop did ______ so I can ______ faster.”

“ Gah，如果Photoshop做到______真好，這樣我就可以______更快。”

Identifying users’ struggles with time-consuming, monotonous, repetitive tasks, or understanding the most common workflows to achieve a particular effect, will help Adobe continue to develop useful machine learning features within each app in the Creative Cloud suite.

識別用戶在費時，單調，重復的任務上的掙扎，或者了解實現特定效果的最常見工作流程，將幫助Adobe繼續在Creative Cloud套件的每個應用程序中開發有用的機器學習功能。