AFAN: Augmented Feature Alignment Network for Cross-Domain Object Detection

被引:25
|
作者
Wang, Hongsong [1 ]
Liao, Shengcai [1 ]
Shao, Ling [1 ]
机构
[1] Incept Inst Artificial Intelligence, Abu Dhabi, U Arab Emirates
关键词
Training; Object detection; Feature extraction; Detectors; Generators; Semantics; Proposals; unsupervised domain adaptation;
D O I
10.1109/TIP.2021.3066046
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Unsupervised domain adaptation for object detection is a challenging problem with many real-world applications. Unfortunately, it has received much less attention than supervised object detection. Models that try to address this task tend to suffer from a shortage of annotated training samples. Moreover, existing methods of feature alignments are not sufficient to learn domain-invariant representations. To address these limitations, we propose a novel augmented feature alignment network (AFAN) which integrates intermediate domain image generation and domain-adversarial training into a unified framework. An intermediate domain image generator is proposed to enhance feature alignments by domain-adversarial training with automatically generated soft domain labels. The synthetic intermediate domain images progressively bridge the domain divergence and augment the annotated source domain training data. A feature pyramid alignment is designed and the corresponding feature discriminator is used to align multi-scale convolutional features of different semantic levels. Last but not least, we introduce a region feature alignment and an instance discriminator to learn domain-invariant features for object proposals. Our approach significantly outperforms the state-of-the-art methods on standard benchmarks for both similar and dissimilar domain adaptations. Further extensive experiments verify the effectiveness of each component and demonstrate that the proposed network can learn domain-invariant representations.
引用
收藏
页码:4046 / 4056
页数:11
相关论文
共 50 条
  • [1] DIVERGENCE-GUIDED FEATURE ALIGNMENT FOR CROSS-DOMAIN OBJECT DETECTION
    Li, Zongyao
    Togo, Ren
    Ogawa, Takahiro
    Haseyama, Miki
    [J]. 2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), 2022, : 2240 - 2244
  • [2] Cross-domain object detection by local to global object-aware feature alignment
    Yiguo Song
    Zhenyu Liu
    Ruining Tang
    Guifang Duan
    Jianrong Tan
    [J]. Neural Computing and Applications, 2024, 36 : 3631 - 3644
  • [3] Cross-domain object detection by local to global object-aware feature alignment
    Song, Yiguo
    Liu, Zhenyu
    Tang, Ruining
    Duan, Guifang
    Tan, Jianrong
    [J]. NEURAL COMPUTING & APPLICATIONS, 2024, 36 (07): : 3631 - 3644
  • [4] SAN: Selective Alignment Network for Cross-Domain Pedestrian Detection
    Jiao, Yifan
    Yao, Hantao
    Xu, Changsheng
    [J]. IEEE TRANSACTIONS ON IMAGE PROCESSING, 2021, 30 : 2155 - 2167
  • [5] HIERARCHICAL DOMAIN-CONSISTENT NETWORK FOR CROSS-DOMAIN OBJECT DETECTION
    Liu, Yuanyuan
    Liu, Ziyang
    Fang, Fang
    Fu, Zhanghua
    Chen, Zhanlong
    [J]. 2021 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), 2021, : 474 - 478
  • [6] A gated cross-domain collaborative network for underwater object detection
    Dai, Linhui
    Liu, Hong
    Song, Pinhao
    Liu, Mengyuan
    [J]. PATTERN RECOGNITION, 2024, 149
  • [7] Decompose to Adapt: Cross-Domain Object Detection Via Feature Disentanglement
    Liu, Dongnan
    Zhang, Chaoyi
    Song, Yang
    Huang, Heng
    Wang, Chenyu
    Barnett, Michael
    Cai, Weidong
    [J]. IEEE TRANSACTIONS ON MULTIMEDIA, 2023, 25 : 1333 - 1344
  • [8] Cross-domain Federated Object Detection
    Su, Shangchao
    Li, Bin
    Zhang, Chengzhi
    Yang, Mingzhao
    Xue, Xiangyang
    [J]. 2023 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO, ICME, 2023, : 1469 - 1474
  • [9] Cross-domain pedestrian detection via feature alignment and image quality assessment
    Yao, Jun
    Guo, Zhilin
    Yu, Junjie
    Yan, Nan
    Wang, Qiong
    Yu, Wei
    [J]. ISCIENCE, 2024, 27 (04)
  • [10] Multi-Level Alignment Network for Cross-Domain Ship Detection
    Xu, Chujie
    Zheng, Xiangtao
    Lu, Xiaoqiang
    [J]. REMOTE SENSING, 2022, 14 (10)