A Deeper Look at Dataset Bias

被引：24

作者：

Tommasi, Tatiana ^{[1
]}

Patricia, Novi ^{[2
,3
]}

Caputo, Barbara ^{[4
]}

Tuytelaars, Tinne ^{[5
]}

机构：

[1] Univ N Carolina, Dept Comp Sci, Chapel Hill, NC 27515 USA

[2] Idiap Res Inst, Martigny, Switzerland

[3] Ecole Polytech Fed Lausanne, CH-1015 Lausanne, Switzerland

[4] Univ Roma La Sapienza, I-00185 Rome, Italy

[5] Katholieke Univ Leuven, ESAT PSI, IMinds, Leuven, Belgium

来源：

PATTERN RECOGNITION, GCPR 2015 | 2015年 / 9358卷

关键词：

D O I：

10.1007/978-3-319-24947-6_42

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

The presence of a bias in each image data collection has recently attracted a lot of attention in the computer vision community showing the limits in generalization of any learning method trained on a specific dataset. At the same time, with the rapid development of deep learning architectures, the activation values of Convolutional Neural Networks (CNN) are emerging as reliable and robust image descriptors. In this paper we propose to verify the potential of the DeCAF features when facing the dataset bias problem. We conduct a series of analyses looking at how existing datasets differ among each other and verifying the performance of existing debiasing methods under different representations. We learn important lessons on which part of the dataset bias problem can be considered solved and which open questions still need to be tackled.

引用

页码：504 / 516

页数：13