Saliency for fine-grained object recognition in domains with scarce training data

被引:41
作者
Figueroa Flores, Carola [1 ,2 ]
Gonzalez-Garcia, Abel [1 ]
van de Weijer, Joost [1 ]
Raducanu, Bogdan [1 ]
机构
[1] Comp Vis Ctr, Edifici O,Campus UAB, Bellaterra 08193, Barcelona, Spain
[2] Univ Bio Bio, Dept Comp Sci & Informat Technol, Concepcion, Chile
关键词
Object recognition; Fine-grained classification; Saliency detection; Scarce training data; VISUAL-ATTENTION;
D O I
10.1016/j.patcog.2019.05.002
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
This paper investigates the role of saliency to improve the classification accuracy of a Convolutional Neural Network (CNN) for the case when scarce training data is available. Our approach consists in adding a saliency branch to an existing CNN architecture which is used to modulate the standard bottom-up visual features from the original image input, acting as an attentional mechanism that guides the feature extraction process. The main aim of the proposed approach is to enable the effective training of a fine-grained recognition model with limited training samples and to improve the performance on the task, thereby alleviating the need to annotate a large dataset. The vast majority of saliency methods are evaluated on their ability to generate saliency maps, and not on their functionality in a complete vision pipeline. Our proposed pipeline allows to evaluate saliency methods for the high-level task of object recognition. We perform extensive experiments on various fine-grained datasets (Flowers, Birds, Cars, and Dogs) under different conditions and show that saliency can considerably improve the network's performance, especially for the case of scarce training data. Furthermore, our experiments show that saliency methods that obtain improved saliency maps (as measured by traditional saliency benchmarks) also translate to saliency methods that yield improved performance gains when applied in an object recognition pipeline. (C) 2019 Published by Elsevier Ltd.
引用
收藏
页码:62 / 73
页数:12
相关论文
共 45 条
  • [1] [Anonymous], ARXIV180606193 CORR
  • [2] [Anonymous], TECHNICAL REPORT
  • [3] [Anonymous], 2017, INT C LEARN REPR
  • [4] [Anonymous], PROC CVPR IEEE
  • [5] [Anonymous], ARXIV161105109 CORR
  • [6] [Anonymous], ARXIV170706484 CORR
  • [7] [Anonymous], 2017, ICLR 2017
  • [8] [Anonymous], 2017, COMMUN ACM, DOI DOI 10.1145/3065386
  • [9] [Anonymous], 2015, INT C COMP VIS ICCV
  • [10] [Anonymous], 2017, P IEEE INT C COMPUTE