Two at Once: Enhancing Learning and Generalization Capacities via IBN-Net

被引:611
作者
Pan, Xingang [1 ]
Luo, Ping [1 ]
Shi, Jianping [2 ]
Tang, Xiaoou [1 ]
机构
[1] Chinese Univ Hong Kong, CUHK SenseTime Joint Lab, Shatin, Hong Kong, Peoples R China
[2] SenseTime Grp Ltd, Beijing, Peoples R China
来源
COMPUTER VISION - ECCV 2018, PT IV | 2018年 / 11208卷
基金
中国国家自然科学基金;
关键词
Instance normalization; Invariance; Generalization; CNNs;
D O I
10.1007/978-3-030-01225-0_29
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Convolutional neural networks (CNNs) have achieved great successes in many computer vision problems. Unlike existing works that designed CNN architectures to improve performance on a single task of a single domain and not generalizable, we present IBN-Net, a novel convolutional architecture, which remarkably enhances a CNN's modeling ability on one domain (e.g. Cityscapes) as well as its generalization capacity on another domain (e.g. GTA5) without finetuning. IBN-Net carefully integrates Instance Normalization (IN) and Batch Normalization (BN) as building blocks, and can be wrapped into many advanced deep networks to improve their performances. This work has three key contributions. (1) By delving into IN and BN, we disclose that IN learns features that are invariant to appearance changes, such as colors, styles, and virtuality/reality, while BN is essential for preserving content related information. (2) IBN-Net can be applied to many advanced deep architectures, such as DenseNet, ResNet, ResNeXt, and SENet, and consistently improve their performance without increasing computational cost. (3) When applying the trained networks to new domains, e.g. from GTA5 to Cityscapes, IBN-Net achieves comparable improvements as domain adaptation methods, even without using data from the target domain. With IBN-Net, we won the 1st place on the WAD 2018 Challenge Drivable Area track, with an mIoU of 86.18%.
引用
收藏
页码:484 / 500
页数:17
相关论文
共 31 条
[1]  
[Anonymous], 2017, COMPUTER VISION PATT
[2]  
[Anonymous], 2017, CoRR
[3]  
[Anonymous], 2017, TPAMI
[4]  
[Anonymous], 2014, ABS14123474 CORR
[5]  
[Anonymous], 2015, arXiv: Learning
[6]  
[Anonymous], 2015, ICML
[7]  
[Anonymous], 2016, FCNS WILD PIXELLEVEL
[8]  
[Anonymous], 2012, ECCV
[9]  
[Anonymous], 2017, A Learned Representation For Artistic Style
[10]  
[Anonymous], 2018, AAAI