Exploring part-aware segmentation for fine-grained visual categorization
- PDF / 5,382,387 Bytes
- 20 Pages / 439.642 x 666.49 pts Page_size
- 26 Downloads / 179 Views
Exploring part-aware segmentation for fine-grained visual categorization Cheng Pang1 · Hongxun Yao1 · Xiaoshuai Sun1 · Sicheng Zhao1 · Yanhao Zhang1
Received: 15 July 2017 / Revised: 27 January 2018 / Accepted: 2 April 2018 © Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract It is challenge to segment fine-grained objects due to appearance variations and clutter of backgrounds. Most of existing segmentation methods hardly separate small parts of the instance from its background with sufficient accuracy. However, such small parts usually contain important semantic information, which is crucial in fine-grained categorization. Observing that fine-grained objects almost share the same configuration of parts, we present a novel part-aware segmentation method, which explicitly detects semantic parts and preserve these parts during segmentation. We firstly design a hybrid part localization method, which generates accurate part proposals with moderate computation. Then we iteratively update the segmentation outputs and the part proposals, which obtains better foreground segmentation results. Experiments demonstrate the superiority of the proposed method, as compared to state-of-the-art segmentation approaches for fine-grained categorization. Keywords Image segmentation · Fine-grained visual categorization · GrabCut
1 Introduction In recent years, fine-grained visual categorization (FGVC) has received increasing attention in computer vision and already been applied in various commercial applications. However, FGVC remains a challenging task due to the following reasons: 1) high inter-class similarities of fine-grained objects and 2) intra-class disparities arising from a large variation of views and object poses. To address these difficulties, recent works are dedicated to learning discriminative features coupled with object alignment for various kinds of fine-grained objects, such as different models of cars [47, 63], breed of animals [4, 18, 28, 41], category of food [5, 64], clothing [36] and actions [19, 46, 55].
Hongxun Yao
[email protected] 1
School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China
Multimed Tools Appl
Although most works on FGVC focus on end-to-end classification, few of them give attention to the benefit of segmentation towards FGVC. Angelova et al. [2] firstly introduced segmentation in FGVC and demonstrated that the accuracy of recognition can be boosted with the help of segmentation. However, most fo existing segmentation methods fail in separating the small parts of the instance from the background with sufficient accuracy. From the perspective of cognitive psychology, basic-level categories are principally defined by their parts while subordinate-level categories are distinguished by the different properties of these parts [44]. As a result, common segmentation methods often lead to poor performance in classification due to the absence of object parts. Based on this observation, we present a novel part-aware segmentation method
Data Loading...
