Seeing AI - Talking Camera App for the Blind|APP|广告营销案例【ADGuider广告指南】

为盲人观看人工智能摄像头应用

案例简介：概要我们的使命是让地球上的每个人取得更多的成就。我们认识到，盲人和低视力社区 (世卫组织称 2.53亿) 往往得不到技术的充分服务，而且历史上教育和就业机会减少了。我们发现了现有辅助工具的三个关键问题: 功能有限: 可用技术感觉非常基本，功能有限或单一用途，如检测颜色或阅读文本。•昂贵: 例如，大多数盲人文本阅读智能眼镜解决方案的价格在 3000-10000 美元之间。同样，为盲人社区设计的基于硬件的激光条形码扫描仪的成本高达 1300 美元。对于一个历来减少教育和就业机会的社区来说，这种成本使得获得这些技术变得更加困难。•笨重的硬件: 由于许多笨重的硬件工具都有一个功能，这意味着在你的包里携带多个物品，降低了流动性。战略通过让用户先于技术，我们开始解决用户任务，然后构建技术来解决它。我们确定了几个场景和感兴趣的对象，并开始收集数据 (通过员工的内部众包)。场景识别模式的训练是通过首先用五个描述性句子标记 300,000 多幅图像 (超过 70,000 小时的人类标记努力)，然后训练管道模型来实现的。该模型识别视觉对象、动作和形容词。然后，它生成一个可能的句子描述，结合对象的上下文以及彼此之间的关系 (例如，桌子上的笔记本电脑而不是笔记本电脑和桌子)。此外，该系统识别名人和地标 (例如 “金门桥的景观” 而不是 “桥”)。该系统帮助一位盲人教授在波多黎各的飓风中避开倒下的电线和倒下的树木，到达安全的地方。结果自 2017年7月推出以来，该应用通过帮助用户独立完成 700万多项任务，提高了独立性，截至 2018年5月，已有 200,000 名用户下载了该应用。许多这些任务以前需要依靠有远见的援助。苹果应用商店上被评为 4.8 名明星，反馈通常包含 “新信心” 、 “新机会” 、 “新体验” 、 “游戏改变者” 、 “独立” 和 “成功” 等词。收到的反馈的一个子集通常围绕以下主题: 生活中的第一次体验 -- 阅读菜单、自拍、在自动售货机中寻找物品教育 -- 学校里的盲人现在可以阅读无法阅读的纸质文本。同样，盲人父母也在看书，检查他们孩子的手写家庭作业和老师的笔记。社会对话中包含的就业感觉盲目摄影杂货店购物和在厨房自力更生获得负担得起的技术。执行凭借独特的直观和创造性的基于音频的用户体验，观看人工智能运行设备上的人工智能，以前所未有的方式解决盲人社区面临的问题。鉴于盲人用户通常很难正确取景相机, 该应用程序使用实时设备上的人工智能来引导他们拍摄更好的照片，从而提高准确性 -- 例如，声音引导，直到文档的所有 4 个角落都可见。同样，为了能够识别条形码，应用程序生成了音频蜂鸣音，以帮助引导用户进入带有条形码的区域。以前，盲人用户必须购买一台价值 1300 美元的专用硬件条形码扫描仪才能做到这一点。当相机在 “人模式” 中看到一张脸时，它描述了这张脸有多远，它在图片中的位置，以帮助用户将它框起来，以及这张脸是微笑还是另一种情绪。有了所有这些声音反馈，盲人用户已经在社交媒体上发布了他们朋友的微笑照片。此外，用户可以拍摄三张照片并训练面部识别器 (这是行业中的第一张)，所以如果一个朋友出现在相机前，它会实时宣布。活动描述观看人工智能是一个免费的应用程序，为盲人和低视力社区讲述视觉世界。这个正在进行的研究项目利用人工智能的力量来打开视觉世界，用语音描述附近的人、文本、货币、颜色和物体。就像瑞士军刀一样，这个应用程序通过利用设备上的深度学习的力量打开许多 “生活中的第一次” 场景，提供了许多工具。主要功能包括-实时文本阅读、文档结构理解、基于音频的条形码定位器和产品识别器、人脸识别和情感/年龄/性别描述、货币识别、颜色识别, 可听光探测器和手写阅读器。此外，这个应用程序可以通过在使用其他应用程序时描述照片来使其他应用程序易于访问和包容。看到 90 秒产品视频: http://aka.ms/seeingaiproductvideo

为盲人观看人工智能摄像头应用

案例简介：Synopsis Our mission is to empower every person on the planet to achieve more. We recognize that the blind and low vision community (253 million according to WHO) is often underserved by technology and historically has reduced educational and employment opportunities. We discovered three key issues with existing accessibility tools: • Limited functionality: Available technologies feel really basic, have limited or single purpose functionality, like detecting color or reading text. • Expensive : For example, most text reading smartglass solutions for the blind are in the range of $3000-$10000. Similarly, hardware-based laser barcode scanners designed for the blind community cost upwards of $1300. For a community which historically has reduced educational and employment opportunities, such costs make it even harder to get access to these technologies. • Bulky hardware: Since many of these bulky hardware tools do one feature, this means carrying multiple items in your bag, reducing mobility. Strategy By keeping the user before the technology, we started addressing the user tasks and then building the technology to solve it. We identified several scenarios and objects of interest and started collecting data (with both internal crowdsourcing from employees). Training for the scene recognition mode was achieved by first labelled over 300,000 images (over 70,000 hours of human labelling effort) with five descriptive sentences each, and then training a pipeline model. This model identifies visual objects, actions, adjectives. Then it generates a possible sentence description, combining the context of objects and relationships between each other (e.g. laptop on top of a table instead of laptop and table). Additionally, the system recognize celebrities, and landmarks (e.g. ‘a view of golden gate bridge’ instead of ‘a bridge’). The system helped a blind professor during the hurricanes in Puerto Rico avoid fallen power lines and fallen trees, and come to safety. Outcome Since launch in July 2017, the app has improved independence by assisting users in completing over 7 million tasks independently and has been downloaded by 200,000 users as of May 2018. Many of these tasks previously required relying on sighted assistance. Rated 4.8 stars on the Apple App Store, feedback often contains words such as ‘new confidence’, ‘new opportunities’, ‘new experiences’, ‘game changer’, ‘’independence’ and ‘empowerment’. A subset of feedback received often revolves around the following themes : First time experiences in life – reading menus, selfies, finding items in a vending machine Education – Blind students in school can now read inaccessible paper text. Similarly, blind parents are reading books, checking the handwritten homework of their kids and notes from teachers. Employment Feeling included in social conversations Blind photography Grocery shopping and self-reliance in the kitchen Getting access to affordable technology. Execution With a uniquely intuitive and creative audio-based user experience, Seeing AI runs on-device AI to solve problems faced by the blind community in a manner that had never been done before. Given that blind users often have difficulty framing camera correctly, the app uses real time AI on-device to guide them to take a better photograph resulting in higher accuracy - e.g. audibly guiding till all 4 corners of a document are visible. Similarly, to be able to recognize barcodes, the app generated audio beeps to help guide the user towards areas with barcodes. Previously, blind users would have to buy a bulky $1300 dedicated hardware barcode scanner to be able to do this. When the camera sees a face in 'person mode', it describes how far the face is, its location in the picture to help users frame it, and if the face is smiling or another emotion. With all this audible feedback, blind users are already posting smiling photographs of their friends on social media. Additionally, users can take three photos and train an face recognizer (a first in the industry), so if a friend appears in front of the camera, it will announce it in real time. CampaignDescription Seeing AI is a free app that narrates the visual world for the blind and low vision community. This ongoing research project harnesses the power of Artificial Intelligence to open up the visual world and describe nearby people, text, currency, color and objects with spoken audio. Much like a Swiss Army knife, this app provides many tools in one by leveraging the power of on-device deep learning to open many “first time in life” scenarios. Key features include - Real-time text reading, document structure understanding, audio-based barcode locator and product recognizer, face recognition and emotion/age/gender description, currency recognition, color recognition, audible light detector and handwriting reader. Additionally, this app can make even other apps accessible and inclusive by describing photos while using other apps. See the 90 second product video : http://aka.ms/seeingaiproductvideo