人工智能培训

搜索

人工智能论文:用于高效视觉跟踪的对抗特征抽样学习(Adversarial Feature Sampling Learning for Efficient Vis

[复制链接]
1379270383 发表于 2018-9-14 09:10:25 | 显示全部楼层 |阅读模式
1379270383 2018-9-14 09:10:25 327 0 显示全部楼层
人工智能论文:用于高效视觉跟踪的对抗特征抽样学习(Adversarial Feature Sampling Learning for Efficient Visual Tracking)逐个检测框架通常包括两个阶段:第一阶段中目标对象周围的图样,并将每个样本分类为第二阶段中的目标对象或背景。基于逐个检测框架的当前普及者通常在第一阶段中将绘图中的样本作为深度卷积网络的输入,这通常导致高计算负担和低运行速度。在本文中,我们提出了一种新的视觉跟踪方法,使用采样深度卷积特征来解决这个问题。仅将目标对象周围的一个裁剪图像输入到设计的深度卷积网络中,并且通过空间双线性采样在网络的特征图上对样本进行采样。此外,生成的对抗网络被集成到我们的网络框架中,以增强正样本并改善跟踪性能。对基准数据集的大量实验表明,所提出的方法实现了与最先进的跟踪器相当的性能,并且有效地加速了基于原始图像采样的跟踪检测跟踪器。
The tracking-by-detection framework usually consist of two stages: drawingsamples around the target object in the first stage and classifying each sampleas the target object or background in the second stage.Current populartrackers based on tracking-by-detection framework typically draw samples in theraw image as the inputs of deep convolution networks in the first stage, whichusually results in high computational burden and low running speed.In thispaper, we propose a new visual tracking method using sampling deepconvolutional features to address this problem.Only one cropped image aroundthe target object is input into the designed deep convolution network and thesamples is sampled on the feature maps of the network by spatial bilinearresampling.In addition, a generative adversarial network is integrated intoour network framework to augment positive samples and improve the trackingperformance.Extensive experiments on benchmark datasets demonstrate that theproposed method achieves a comparable performance to state-of-the-art trackersand accelerates tracking-by-detection trackers based on raw-image sampleseffectively.人工智能论文:用于高效视觉跟踪的对抗特征抽样学习(Adversarial Feature Sampling Learning for Efficient Visual Tracking) cm6GHpNgd6jjjjLh.jpg
URL地址:https://arxiv.org/abs/1809.04741     ----pdf下载地址:https://arxiv.org/pdf/1809.04741    ----人工智能论文:用于高效视觉跟踪的对抗特征抽样学习(Adversarial Feature Sampling Learning for Efficient Visual Tracking)
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

1379270383当前离线
新手上路

查看:327 | 回复:0

快速回复 返回顶部 返回列表