人工智能培训

搜索

人工智能论文:学习用单一RGB相机重建服装人物(Learning to Reconstruct People in Clothing from a Single

[复制链接]
baiselong 发表于 2019-3-15 12:14:23 | 显示全部楼层 |阅读模式
baiselong 2019-3-15 12:14:23 426 0 显示全部楼层
人工智能论文:学习用单一RGB相机重建服装人物(Learning to Reconstruct People in Clothing from a Single RGB Camera)我们提出了一种基于学习的模型,可以在不到10秒的时间内从人眼正在移动的单眼视频的几帧(1-8)中推断出个性化的3D形状,重建精度为5mm。我们的模型学会预测统计身体模型的参数以及将衣服和头发添加到形状的实例位移。该模型基于两个关键设计选择实现快速准确的预测。首先,通过在规范的T-姿势空间中预测形状,网络学习将人的图像编码为姿势不变的潜在代码,其中信息被融合。其次,基于观察到前馈预测快但不总是与输入图像对齐,我们预测使用自下而上和自下而上的流(每个视图一个),允许信息在两个方向上流动。学习仅依赖于合成3D数据。一旦学会了,模型可以采用可变数量的帧作为输入,并且能够甚至从单个图像重建形状,精度为6mm。 3种不同数据集的结果证明了我们方法的有效性和准确性。
We present a learning-based model to infer the personalized 3D shape ofpeople from a few frames (1-8) of a monocular video in which the person ismoving, in less than 10 seconds with a reconstruction accuracy of 5mm.Ourmodel learns to predict the parameters of a statistical body model and instancedisplacements that add clothing and hair to the shape.The model achieves fastand accurate predictions based on two key design choices.First, by predictingshape in a canonical T-pose space, the network learns to encode the images ofthe person into pose-invariant latent codes, where the information is fused.Second, based on the observation that feed-forward predictions are fast but donot alwaysalign with the input images, we predict using both, bottom-up andtop-down streams (one per view) allowing information to flow in bothdirections.Learning relies only on synthetic 3D data.Once learned, the modelcan take a variable number of frames as input, and is able to reconstructshapes even from a single image with an accuracy of 6mm.Results on 3 differentdatasets demonstrate the efficacy and accuracy of our approach.人工智能论文:学习用单一RGB相机重建服装人物(Learning to Reconstruct People in Clothing from a Single RGB Camera) poR6Kp5Aq7p6hRRb.jpg
URL地址:https://arxiv.org/abs/1903.05885     ----pdf下载地址:https://arxiv.org/pdf/1903.05885    ----人工智能论文:学习用单一RGB相机重建服装人物(Learning to Reconstruct People in Clothing from a Single RGB Camera)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

baiselong当前离线
新手上路

查看:426 | 回复:0

快速回复 返回顶部 返回列表