POCO:用于3D人体姿势和形状估计的新型人工智能框架
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">要点:</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">经过</span>2D图像估计3D人体姿态和形状是一个<span style="color: black;">拥有</span>挑战性的任务,<span style="color: black;">由于</span>存在深度模糊、遮挡和不寻常的服装等问题。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">传统<span style="color: black;">办法</span>在估计不确定性时存在<span style="color: black;">有些</span>限制,而POCO框架<span style="color: black;">供给</span>了一种改进<span style="color: black;">办法</span>,<span style="color: black;">能够</span>在单个前向传递中<span style="color: black;">同期</span>推断姿态参数和不确定性。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">POCO框架引入了<span style="color: black;">要求</span>向量和图像特征来<span style="color: black;">加强</span>基本密度函数的建模,<span style="color: black;">同期</span><span style="color: black;">经过</span>SMPL姿势来<span style="color: black;">调节</span>网络,从而<span style="color: black;">加强</span>了姿态重建和不确定性估计的准确性。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">站长之家(ChinaZ.com)10月16日 <span style="color: black;">信息</span>:</strong>人体姿态和形状(HPS)的三维估计是重建现实世界中的人体<span style="color: black;">行径</span>所必需的。然而,从二维图像进行三维推断面临深度模糊、遮挡、不寻常的服装和运动模糊等挑战。即使最先进的HPS<span style="color: black;">办法</span><span style="color: black;">亦</span>会<span style="color: black;">显现</span>错误,<span style="color: black;">一般</span>对这些错误不太<span style="color: black;">认识</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">HPS是一个中间任务,<span style="color: black;">供给</span>了下游任务所需的输出,如理解人类<span style="color: black;">行径</span>或三维图形应用。这些下游任务需要一种机制来<span style="color: black;">评定</span>HPS结果的准确性,<span style="color: black;">因此呢</span>这些<span style="color: black;">办法</span>必须生成与HPS质量<span style="color: black;">关联</span>的不确定性(或置信度)值。</p><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-tjoges91tu/3c540bb65ea5a9501d3621e33436f431~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1729492145&x-signature=%2BSUqlfBtG4UklDPRkrOv2bXBpr4%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">项目<span style="color: black;">位置</span></p>:https://poco.is.tue.mpg.de/
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">POCO框架的核心创新是"双重<span style="color: black;">要求</span>策略(DCS)",它<span style="color: black;">加强</span>了基本密度函数和规模网络。与先前的<span style="color: black;">办法</span><span style="color: black;">区别</span>,POCO引入了一个<span style="color: black;">要求</span>向量(Cond-bDF)来建模推断的姿态误差的基本密度函数。POCO利用图像特征进行<span style="color: black;">要求</span>化,使其能够更好地适应多样性和<span style="color: black;">繁杂</span>的图像数据集的训练。</p><img src="https://p26-sign.toutiaoimg.com/tos-cn-i-tjoges91tu/dce34ff88a45d58d019ef1136f5cc3d9~noop.image?_iz=58558&from=article.pc_detail&lk3s=953192f4&x-expires=1729492145&x-signature=hmj9MViTXteDovo8CjhBpJAoqkA%3D" style="width: 50%; margin-bottom: 20px;">
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">另外</span>,POCO的作者们引入了一种改进的<span style="color: black;">办法</span>,用于在HPS模型中估计不确定性。<span style="color: black;">她们</span>利用图像特征并将网络与SMPL姿态相结合,从而<span style="color: black;">加强</span>了姿态重建和更好的不确定性估计。<span style="color: black;">她们</span>的<span style="color: black;">办法</span><span style="color: black;">能够</span>无缝集成到现有的HPS模型中,<span style="color: black;">加强</span>准确性而<span style="color: black;">无</span>副<span style="color: black;">功效</span>。该<span style="color: black;">科研</span>声<span style="color: black;">叫作</span>这种<span style="color: black;">办法</span>在将不确定性与姿态错误<span style="color: black;">关联</span>方面优于最先进的<span style="color: black;">办法</span>。</p>
<p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">POCO框架是一个创新的AI工具,用于三维人体姿态和形状的估计。它<span style="color: black;">经过</span>引入双重<span style="color: black;">要求</span>策略,改进基本密度函数和规模网络,<span style="color: black;">明显</span><span style="color: black;">加强</span>了估计的准确性。这一<span style="color: black;">办法</span><span style="color: black;">能够</span>应用于多样性和<span style="color: black;">繁杂</span>图像数据集,使其<span style="color: black;">作为</span>一种有望推动人体<span style="color: black;">行径</span>分析和三维图形应用的关键技术。<span style="color: black;">倘若</span>您对此感兴趣,不妨查看<span style="color: black;">关联</span>链接以<span style="color: black;">认识</span><span style="color: black;">更加多</span>详情。</p>
页:
[1]