人工智能培训

搜索

机器学习论文:LPM:用于高效全面凝视估计的可学习池模块(LPM: Learnable Pooling Module for Efficient Full-Fa

[复制链接]
rrrrrrr 发表于 2019-3-15 12:46:10 | 显示全部楼层 |阅读模式
rrrrrrr 2019-3-15 12:46:10 571 0 显示全部楼层
机器学习论文:LPM:用于高效全面凝视估计的可学习池模块(LPM: Learnable Pooling Module for Efficient Full-Face Gaze Estimation)凝视跟踪是许多领域的重要技术。诸如卷积神经网络(CNN)之类的技术已经允许发明凝视跟踪方法,该方法仅依赖于诸如个人计算机上的相机之类的商品硬件。已经表明,用于凝视估计的全脸区域可以提供比仅从眼睛图像更好的性能。然而,使用全脸图像的问题是由于较大的图像尺寸而导致的繁重计算。本研究通过使用新颖的可学习池模块去除冗余信息来压缩输入全脸图像来解决这个问题。可以通过反向传播对模块进行端到端训练,从而允许池化过滤器中网格的大小。可学习的池模块将有价值区域的分辨率保持在高水平,反之亦然。当图像缩小到较小尺寸时,该提出的方法将凝视估计精度保持在一定水平。
Gaze tracking is an important technology in many domains.Techniques such asConvolutional Neural Networks (CNN) has allowed the invention of gaze trackingmethod that relies only on commodity hardware such as the camera on a personalcomputer.It has been shown that the full-face region for gaze estimation canprovide better performance than from an eye image alone.However, a problemwith using the full-face image is the heavy computation due to the larger imagesize.This study tackles this problem through compression of the inputfull-face image by removing redundant information using a novel learnablepooling module.The module can be trained end-to-end by backpropagation tolearn the size of the grid in the pooling filter.The learnable pooling modulekeeps the resolution of valuable regions high and vice versa.This proposedmethod preserved the gaze estimation accuracy at a certain level when the imagewas reduced to a smaller size.机器学习论文:LPM:用于高效全面凝视估计的可学习池模块(LPM: Learnable Pooling Module for Efficient Full-Face Gaze Estimation) NS6Ig9X2X9u9UiJU.jpg
URL地址:https://arxiv.org/abs/1903.05761     ----pdf下载地址:https://arxiv.org/pdf/1903.05761    ----机器学习论文:LPM:用于高效全面凝视估计的可学习池模块(LPM: Learnable Pooling Module for Efficient Full-Face Gaze Estimation)
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

rrrrrrr当前离线
新手上路

查看:571 | 回复:0

快速回复 返回顶部 返回列表