视觉属性学习应用研究

Application Research of Visual Attribute Learning

视觉属性能够展现人们识别事物时所定义的语义概念,但现有的视觉识别工作往往忽略了属性在分类器设计中能作为中间媒介层的作用.本文给出一种将属性和事物类别同时用于构建分类器的方法.分析了传统多类分类器、直接类别相关模型、间接属性预测模型、直接属性预测模型的特点.在动物和室外场景数据集上的实验证明,利用属性进行分类器学习对于提高传统多目标分类和迁移学习的性能都具有很好的帮助.

Visual attributes expose human-defined semantics to object recognition models,but existing work largely restricts their influence to mid-level cues during classifier training.A kind of classifier constructed by both attributes and categories is put forward in this paper.The paper analysis the traditional multi-class classifier,direct related categories model,indirect attribute prediction model and direct attribute prediction.Experiments in animal and outdoor scenes data set shows that attributes have a good help to improve the performance of traditional multi target classification and transfer learning.