谷歌推出史上最强人脸识别系统
<p>“<span word="I">I</span> <span word="never">never</span> <span word="forget">forget</span> <span word="a">a</span> <span word="face">face</span>,” <span word="some">some</span> <span word="people">people</span> <span word="like">like</span> <span word="to">to</span> <span word="boast">boast</span>. <span word="It">It</span>’<span word="s">s</span> <span word="a">a</span> <span word="claim">claim</span> <span word="that">that</span> <span word="looks">looks</span> <span word="quainter">quainter</span> <span word="by">by</span> <span word="the">the</span> <span word="day">day</span> <span word="as">as</span> <span word="artificial">artificial</span> <span word="intelligence">intelligence</span> <span word="research">research</span> <span word="continues">continues</span> <span word="to">to</span> <span word="advance">advance</span>. <span word="Some">Some</span> <span word="computers">computers</span>, <span word="it">it</span> <span word="turns">turns</span> <span word="out">out</span>, <span word="never">never</span> <span word="forget">forget</span> 260 <span word="million">million</span> <span word="faces">faces</span>.</p><p>有些人总喜欢夸口说：“我从来不会忘记别人的长相。”在人工智能研究突飞猛进的今天，还要这么夸口就有点奇怪了。事实上，现在有些电脑能记住2.6亿张脸。</p><p><span word="Last">Last</span> <span word="week">week</span>, <span word="a">a</span> <span word="trio">trio</span> <span word="of">of</span> <span word="Google">Google</span> <span word="GOOG">GOOG</span> -0.66% <span word="researchers">researchers</span> <span word="published">published</span> <span word="a">a</span> <span word="paper">paper</span> <span word="on">on</span> <span word="a">a</span> <span word="new">new</span> <span word="artificial">artificial</span> <span word="intelligence">intelligence</span> <span word="system">system</span> <span word="dubbed">dubbed</span> <span word="FaceNet">FaceNet</span> <span word="that">that</span> <span word="it">it</span> <span word="claims">claims</span> <span word="represents">represents</span> <span word="the">the</span> <span word="most">most</span>-<span word="accurate">accurate</span> <span word="approach">approach</span> <span word="yet">yet</span> <span word="to">to</span> <span word="recognizing">recognizing</span> <span word="human">human</span> <span word="faces">faces</span>. <span word="FaceNet">FaceNet</span> <span word="achieved">achieved</span> <span word="nearly">nearly</span> 100-<span word="percent">percent</span> <span word="accuracy">accuracy</span> <span word="on">on</span> <span word="a">a</span> <span word="popular">popular</span> <span word="facial">facial</span>-<span word="recognition">recognition</span> <span word="dataset">dataset</span> <span word="called">called</span> <span word="Labeled">Labeled</span> <span word="Faces">Faces</span> <span word="in">in</span> <span word="the">the</span> <span word="Wild">Wild</span>, <span word="which">which</span> <span word="includes">includes</span> <span word="more">more</span> <span word="than">than</span> 13,000 <span word="pictures">pictures</span> <span word="of">of</span> <span word="faces">faces</span> <span word="from">from</span> <span word="across">across</span> <span word="the">the</span> <span word="web">web</span>. <span word="Trained">Trained</span> <span word="on">on</span> <span word="a">a</span> <span word="massive">massive</span> 260-<span word="million">million</span>-<span word="image">image</span> <span word="dataset">dataset</span>, <span word="FaceNet">FaceNet</span> <span word="performed">performed</span> <span word="with">with</span> <span word="better">better</span> <span word="than">than</span> 86 <span word="percent">percent</span> <span word="accuracy">accuracy</span>.</p><p>上周，谷歌公司的三位研究人员发表了一篇有关全新人工智能系统的研究论文。这一系统名为<span word="FaceNet">FaceNet</span>，谷歌号称它是迄今为止最精确的人脸识别技术。面对一个名为“人面数据库”（<span word="Labeled">Labeled</span> <span word="Faces">Faces</span> <span word="in">in</span> <span word="the">the</span> <span word="Wild">Wild</span>）的常用人脸识别数据库时，<span word="FaceNet">FaceNet</span>识别的准确率近乎百分之百。这个数据库容纳了网上搜集的一万三千多张人脸照片。而在面对一个含有2.6亿张人脸照片的庞大数据库时，这个系统的准确率也超过86%。</p>
页:
[1]