Facebook打造第一視角視頻數據集Ego4D:捕獲超3000小時鏡頭,劍指下一代AI

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據瞭解,Ego4D 是目前最大的第一視角日常活動視頻數據集。"}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"第一視角視頻數據集 Ego4D"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"近日,Facebook 公佈了一項名爲 "},{"type":"link","attrs":{"href":"https:\/\/ego4d-data.org\/","title":"xxx","type":null},"content":[{"type":"text","text":"Ego4D"}]},{"type":"text","text":" 的研究項目。該項目爲 Facebook 與全球 13 所大學和實驗室合作項目,通過收集第一人稱鏡頭,以訓練下一代人工智能模型。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"信息顯示,Ego4D 數據集包含"},{"type":"link","attrs":{"href":"https:\/\/arxiv.org\/abs\/2110.07058","title":"xxx","type":null},"content":[{"type":"text","text":"超過 3025 個小時的視頻"}]},{"type":"text","text":",由來自 9 個國家(美國、英國、印度、日本、意大利、新加坡、沙特阿拉伯、哥倫比亞和盧旺達)73 個不同地點錄製的視頻組成,總錄製人數達 855 人。據瞭解,這些參與者擁有不同的年齡和背景,有些人是因其有趣的職業而被招募過來,例如麪包師、機械師、木匠和園藝師。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/resource\/image\/6d\/5a\/6d1c21f9eb29f19cfe25ff99dd4ec75a.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這也是目前最大的"},{"type":"link","attrs":{"href":"https:\/\/www.cnbc.com\/2021\/10\/14\/facebook-announces-ego4d-first-person-video-data-set-for-training-ai.html","title":"xxx","type":null},"content":[{"type":"text","text":"第一視角日常活動視頻數據集"}]},{"type":"text","text":",在此之前,最大的第一視角視頻數據集由人在廚房裏 100 個小時的鏡頭組成。此外,以前的數據集通常由只有幾秒鐘的半腳本視頻剪輯組成,而 Ego4D 的參與者一次佩戴頭戴式攝像頭長達 10 小時,並拍攝無腳本日常活動的第一人稱視頻,包括沿街散步、閱讀、洗衣、購物、與寵物玩耍、玩棋盤遊戲和與其他人互動。一些鏡頭還包括音頻、有關參與者注視焦點位置的數據以及同一場景的多個視角。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/resource\/image\/eb\/77\/ebe4abca3f7b1688e6d7998d423a9e77.jpeg","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"收集到視頻後,盧旺達的工作人員總共花費了 25 萬個小時觀看數千個視頻剪輯,並編寫數百萬個描述拍攝場景和活動的句子。這些視頻能夠幫助人工智能理解或識別現實世界或虛擬世界中的某些事物,人類也可以通過一副眼鏡或 Oculus 耳機從第一人稱視角看到這些事物。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"紐約石溪大學和谷歌大腦的計算機視覺研究員 Michael Ryoo 表示:“這個數據集裏的視頻更接近人類所觀察的世界,這在同類數據集中是第一個。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"值得一提的是,研究人員還列出了該項目的五大挑戰:"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"情景記憶:我的 X 在哪裏?"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"手與物體交互:物體在交互過程中如何變化?"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"視聽日記:誰說了什麼,什麼時候說的?"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"社會交互:誰在與誰交互?"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"預測:接下來會發生什麼?"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Facebook 首席研究科學家 Kristen Grauman 在接受 CNBC 採訪時表示,“這次發佈的是一個開放數據集和研究挑戰,它能促進我們內部和學術界外部進步,其他研究人員可以支持這些新問題,以更有意義、更大規模的方式共同解決它”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據 Grauman 介紹,該數據集可以部署在 AI 模型中,用於訓練機器人等技術以更快地瞭解世界。“在過去,機器人通過在自己做事來進行學習,現在,它們有機會根據人類經驗從視頻中學習。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Facebook 表示,Ego4D 數據集將在 2021 年 11 月底之前提供下載。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"隱私問題引擔憂"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Ego4D 數據集雖然給下一代人工智能帶來了更多的想象空間,但也不可避免地引發人們對於"},{"type":"link","attrs":{"href":"https:\/\/www.technologyreview.com\/2021\/10\/14\/1037043\/facebook-machine-learning-ai-vision-see-world-human-eyes\/","title":"xxx","type":null},"content":[{"type":"text","text":"隱私問題"}]},{"type":"text","text":"的擔憂。Grauman 坦言:“在做 Ego4D 項目時,我們也意識到有一些隱私方面的工作需要做,尤其是當將隱私從探索性研究領域帶出融入到產品中時。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Facebook 表示,只有在徵得參與者同意後,數據纔會包含面部和其他識別信息。同時,出於隱私考慮,對於大多數視頻,數據已在發佈前進行了去標識化處理,如已從視頻中刪除了個人身份信息,並模糊了旁觀者的面部和車牌號碼,此外,許多視頻中的音頻也被刪除了。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"面對質疑,Facebook 的發言人稱,該公司預計將來會進一步引入隱私保護措施,“Ego4D 純粹是爲了促進更廣泛科學界進步的研究,我們今天沒有任何關於產品應用或商業用途的分享。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"伴隨着人工智能技術快速發展,隱私問題一直是大家討論的焦點。人工智能在獲取和處理海量信息數據,不可避免會涉及個人隱私保護這一重要倫理問題,並且隱藏着不容忽視的隱私泄露風險。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"當前,國內外都頒佈了相應的法規保護用戶隱私與數據安全。比如在歐洲,2018 年生效的《通用數據保護條例》(General Data Protection Regulation,GDPR)對個人數據的收集和使用進行了規範。數據保護條例並沒有明確提及人工智能或機器學習,但對個人數據的大規模自動處理和自動決策非常重視。這意味着,凡是人工智能使用個人數據的地方,都屬於該條例的範圍,皆適用 GDPR 原則。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而至於 Facebook 的 Ego4D 數據集未來會在隱私保護上交出怎樣的答卷,一切交給時間。"}]}]}
發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章