Discriminative Features Matter: Multi-layer Bilinear Pooling for Camera Localization

Xin Wang; Xiang Wang; Chen  Wang; Xiao Bai; Jing Wu; Edwin R Hancock

Discriminative Features Matter: Multi-layer Bilinear Pooling for Camera Localization

Xin Wang, Xiang Wang, Chen Wang, Xiao Bai, Jing Wu, Edwin R Hancock

Computer Science

Research output: Contribution to conference › Paper › peer-review

Abstract

Deep learning based camera localization from a single image has been explored recently since these methods are computationally efficient. However, existing methods only provide general global representations, from which an accurate pose estimation can not be reliably derived. We claim that effective feature representations for accurate pose estimation shall be both "informative" (focusing on geometrically meaningful regions) and "discriminative" (accounting for different poses of similar images). Therefore, we propose a novel multi-layer factorized bilinear pooling module for feature aggregation. Specifically, informative features are selected via bilinear pooling, and discriminative features are highlighted via multi-layer fusion. We develop a new network for camera localization using the proposed feature pooling module. The effectiveness of our approach is demonstrated by experiments on an outdoor Cambridge Landmarks dataset and an indoor 7 Scenes dataset. The results show that focusing on discriminative features significantly improves the network performance of camera localization in most cases. Codes will be available soon.

Original language	English
Number of pages	12
Publication status	Accepted/In press - 23 Jul 2019
Event	British Machine Vision Conference - Cardiff, Cardiff, United Kingdom Duration: 9 Sept 2019 → 12 Sept 2019 https://bmvc2019.org/

Conference

Conference	British Machine Vision Conference
Abbreviated title	BMVC 2019
Country/Territory	United Kingdom
City	Cardiff
Period	9/09/19 → 12/09/19
Internet address	https://bmvc2019.org/

Bibliographical note

Access to Document

0968 (1)
© 2019. The copyright of this document resides with its authors

Cite this

@conference{b9ee81f838874fe4a22661fb3ee63dd9,

title = "Discriminative Features Matter: Multi-layer Bilinear Pooling for Camera Localization",

abstract = "Deep learning based camera localization from a single image has been explored recently since these methods are computationally efficient. However, existing methods only provide general global representations, from which an accurate pose estimation can not be reliably derived. We claim that effective feature representations for accurate pose estimation shall be both {"}informative{"} (focusing on geometrically meaningful regions) and {"}discriminative{"} (accounting for different poses of similar images). Therefore, we propose a novel multi-layer factorized bilinear pooling module for feature aggregation. Specifically, informative features are selected via bilinear pooling, and discriminative features are highlighted via multi-layer fusion. We develop a new network for camera localization using the proposed feature pooling module. The effectiveness of our approach is demonstrated by experiments on an outdoor Cambridge Landmarks dataset and an indoor 7 Scenes dataset. The results show that focusing on discriminative features significantly improves the network performance of camera localization in most cases. Codes will be available soon.",

author = "Xin Wang and Xiang Wang and Chen Wang and Xiao Bai and Jing Wu and Hancock, {Edwin R}",

note = "{\textcopyright} 2019. The copyright of this document resides with its authors; British Machine Vision Conference, BMVC 2019 ; Conference date: 09-09-2019 Through 12-09-2019",

year = "2019",

month = jul,

day = "23",

language = "English",

url = "https://bmvc2019.org/",

}

TY - CONF

T1 - Discriminative Features Matter: Multi-layer Bilinear Pooling for Camera Localization

AU - Wang, Xin

AU - Wang, Xiang

AU - Wang, Chen

AU - Bai, Xiao

AU - Wu, Jing

AU - Hancock, Edwin R

PY - 2019/7/23

Y1 - 2019/7/23

N2 - Deep learning based camera localization from a single image has been explored recently since these methods are computationally efficient. However, existing methods only provide general global representations, from which an accurate pose estimation can not be reliably derived. We claim that effective feature representations for accurate pose estimation shall be both "informative" (focusing on geometrically meaningful regions) and "discriminative" (accounting for different poses of similar images). Therefore, we propose a novel multi-layer factorized bilinear pooling module for feature aggregation. Specifically, informative features are selected via bilinear pooling, and discriminative features are highlighted via multi-layer fusion. We develop a new network for camera localization using the proposed feature pooling module. The effectiveness of our approach is demonstrated by experiments on an outdoor Cambridge Landmarks dataset and an indoor 7 Scenes dataset. The results show that focusing on discriminative features significantly improves the network performance of camera localization in most cases. Codes will be available soon.

AB - Deep learning based camera localization from a single image has been explored recently since these methods are computationally efficient. However, existing methods only provide general global representations, from which an accurate pose estimation can not be reliably derived. We claim that effective feature representations for accurate pose estimation shall be both "informative" (focusing on geometrically meaningful regions) and "discriminative" (accounting for different poses of similar images). Therefore, we propose a novel multi-layer factorized bilinear pooling module for feature aggregation. Specifically, informative features are selected via bilinear pooling, and discriminative features are highlighted via multi-layer fusion. We develop a new network for camera localization using the proposed feature pooling module. The effectiveness of our approach is demonstrated by experiments on an outdoor Cambridge Landmarks dataset and an indoor 7 Scenes dataset. The results show that focusing on discriminative features significantly improves the network performance of camera localization in most cases. Codes will be available soon.

M3 - Paper

T2 - British Machine Vision Conference

Y2 - 9 September 2019 through 12 September 2019

ER -