20170712-14 赵恒爽：PSPNet and ICNet: Semantic Segmentation with...

报告嘉宾：赵恒爽 (The Chinese University of Hong Kong)

报告时间：2017年7月12日（星期三）晚上21:00

报告题目：PSPNet and ICNet: Semantic Segmentation with High Accuracy and High Efficiency

主持人：郭裕兰（国防科技大学）

报告摘要：

In this talk, I will introduce our focuses on two major criteria in semantic segmentation task, the inference performance on accuracy and efficiency aspects. First, to build a high accuracy framework, we exploit the capability of global context information by different-region-based context aggregation through our pyramid pooling module, the proposed pyramid scene parsing network (PSPNet) is effective to deal with unrestricted open vocabulary and diverse scenes. Second, to establish a high efficiency system, we propose image cascade network (ICNet), it incorporates multi-resolution branches under proper label guidance together with cascade feature fusion unit to speed up the network prediction. PSPNet sets up new records on various datasets and it won first place in ImageNet scene parsing challenge 2016, ICNet yields real-time inference speed on high resolution images which is beneficial for many practical applications.

参考文献：

[1] Pyramid Scene Parsing Network. Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, Jiaya Jia. CVPR 2017.

[2] ICNet for Real-Time Semantic Segmentation on High-Resolution Images. Hengshuang Zhao, Xiaojuan Qi, Xiaoyong Shen, Jianping Shi, Jiaya Jia. arXiv:1704.08545, 2017.

报告人简介：

Hengshuang Zhao is currently a second year Ph.D. student in Computer Science at The Chinese University of Hong Kong, supervised by Prof. Jiaya Jia. His research interests lie in computer vision and deep learning. He was a recipient of Microsoft-IEEE Young Fellow Scholarship Award. He and his team won 1st place in ImageNet Scene Parsing Challenge 2016.

特别鸣谢本次Webinar主要组织者：

VOOC责任委员：郭裕兰（国防科技大学）

VODB协调理事：邓成（西安电子科技大学）