An Improved Neural Network Model Based on Visual Attention Mechanism for Object Detection

DOI

10.2991/acsr.k.191223.035How to use a DOI?

Keywords

object detection, cornernet, visual attention mechanism, inference time

Abstract

The general object detection methods include one-stage and two-stage object detection algorithm. The two-stage approach, such as R-CNN family, is composed by the RPN network and object classification network with a better accuracy. The one-stage object detection algorithm represented by YOLO and CornerNet, which are end-to-end structure. This paper proposes an improved CornerNet structure with soft-attention mechanism, which increases the attention weight in the corresponding corner prediction parts of the hourglass model to compensate visually under occlusion or weak light condition. Experiments based on MS COCO dataset show that the proposed structure can lower the inference time further with basically unchanged mAP under the same conditions.

Copyright

© 2019, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Zeren Jiang
PY  - 2019
DA  - 2019/12/24
TI  - An Improved Neural Network Model Based on Visual Attention Mechanism for Object Detection
BT  - Proceedings of the 2019 International Conference on Big Data, Electronics and Communication Engineering (BDECE 2019)
PB  - Atlantis Press
SP  - 149
EP  - 152
SN  - 2352-538X
UR  - https://doi.org/10.2991/acsr.k.191223.035
DO  - 10.2991/acsr.k.191223.035
ID  - Jiang2019
ER  -