DeepMind已开发具有三维想象力的视觉计算机
<p><span word="DeepMind">DeepMind</span>, <span word="Google">Google</span>'<span word="s">s</span> <span word="artificial">artificial</span> <span word="intelligence">intelligence</span> <span word="subsidiary">subsidiary</span> <span word="in">in</span> <span word="London">London</span>, <span word="has">has</span> <span word="developed">developed</span> <span word="a">a</span> <span word="self">self</span>-<span word="training">training</span> <span word="vision">vision</span> <span word="computer">computer</span> <span word="that">that</span> <span word="generates">generates</span> "<span word="a">a</span> <span word="full">full</span> 3<span word="D">D</span> <span word="model">model</span> <span word="of">of</span> <span word="a">a</span> <span word="scene">scene</span> <span word="from">from</span> <span word="just">just</span> <span word="a">a</span> <span word="handful">handful</span> <span word="of">of</span> 2<span word="D">D</span> <span word="snapshots">snapshots</span>", <span word="according">according</span> <span word="to">to</span> <span word="its">its</span> <span word="chief">chief</span> <span word="executive">executive</span>.</p><p>位于伦敦的谷歌人工智能子公司<span word="DeepMind">DeepMind</span>，近日开发了一款自我训练的视觉计算机。据其首席执行官介绍，这款计算机“仅利用几张2<span word="D">D</span>快照就能生成一个完整的3<span word="D">D</span>场景模型”。</p><p><span word="The">The</span> <span word="system">system</span>, <span word="called">called</span> <span word="the">the</span> <span word="Generative">Generative</span> <span word="Query">Query</span> <span word="Network">Network</span>, <span word="can">can</span> <span word="then">then</span> <span word="imagine">imagine</span> <span word="and">and</span> <span word="render">render</span> <span word="the">the</span> <span word="scene">scene</span> <span word="from">from</span> <span word="any">any</span> <span word="angle">angle</span>, <span word="said">said</span> <span word="Demis">Demis</span> <span word="Hassabis">Hassabis</span>.</p><p>杰米斯·哈萨比斯表示，这套被称为“生成式查询网络”的系统可以从任何角度想象和呈现场景。</p>
页:
[1]