| 基于反向投影的zero-shot Learning目标分类算法研究 |
| |
| 引用本文： | 冯 鹏,庹红娅,乔凌峰,王洁欣,敬忠良.基于反向投影的zero-shot Learning目标分类算法研究[J].计算机应用研究,2017,34(11). |
| |
| 作者姓名： | 冯 鹏 庹红娅 乔凌峰 王洁欣 敬忠良 |
| |
| 作者单位： | 上海交通大学航空航天学院,上海交通大学航空航天学院,上海交通大学航空航天学院,上海交通大学航空航天学院,上海交通大学航空航天学院 |
| |
| 基金项目： | 国家自然科学基金,61175028。 |
| |
| 摘 要： | zero-shot learning是对没有训练样本的类别进行分类的问题。传统回归方法的核心是将视觉特征投影到语义空间,没有充分利用视觉特征自身包含的样本信息,同时训练计算量大。本文提出基于反向投影的ZSL目标分类方法,将类别原型投影到视觉空间,利用视觉特征的语义性学习出映射函数,参数优化过程仅通过解析解就可以获得。在两个基准数据集的实验结果表明,我们提出的反向投影方法分类结果较传统回归方法和其他现有方法有大幅提升,且训练时间大大减少,我们的方法可以更好推广到未知类别的分类问题上。
|
| 关 键 词： | zero-shot learning 目标分类 反向投影 解析解 |
| 收稿时间： | 2016/8/3 0:00:00 |
| 修稿时间： | 2017/8/29 0:00:00 |
Zero-shot object recognition based on inverse projection |
| |
| fengpeng,tuohongy,qiaolingfeng,wangjiexin and jingzhongliang.Zero-shot object recognition based on inverse projection[J].Application Research of Computers,2017,34(11). |
| |
| Authors: | fengpeng tuohongy qiaolingfeng wangjiexin and jingzhongliang |
| |
| Affiliation: | School of Aeronautics and Astronautics,Shanghai Jiaotong University,China,School of Aeronautics and Astronautics,Shanghai Jiaotong University,China,School of Aeronautics and Astronautics,Shanghai Jiaotong University,China,School of Aeronautics and Astronautics,Shanghai Jiaotong University,China,School of Aeronautics and Astronautics,Shanghai Jiaotong University,China |
| |
| Abstract: | Zero-shot learning (ZSL) aims to recognise new objects without having training samples of them.In this paper,we propose to project prototype to visual feature space,which is referred to as inverse projection.It has a very efficient closed-form solution. Extensive experiments on two benchmark datasets show that our ZSL method significantly outperforms the state-of-the-arts. |
| |
| Keywords: | zero-shot learning objects recognition inverse projection closed-form solution |
|
| 点击此处可从《计算机应用研究》浏览原始摘要信息 |
|
点击此处可从《计算机应用研究》下载全文 |
|
