Cross-domain pedestrian detection via feature alignment and image quality assessment
Summary: Datasets collected under different sensors, viewpoints, or weather conditions cause different domains. Models trained on domain A applied to tasks of domain B result in low performance. To overcome the domain shift, we propose an unsupervised pedestrian detection method that utilizes CycleG...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2024-04-01
|
| Series: | iScience |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2589004224008617 |
| _version_ | 1797213358253932544 |
|---|---|
| author | Jun Yao Zhilin Guo JunJie Yu Nan Yan Qiong Wang Wei Yu |
| author_facet | Jun Yao Zhilin Guo JunJie Yu Nan Yan Qiong Wang Wei Yu |
| author_sort | Jun Yao |
| collection | DOAJ |
| description | Summary: Datasets collected under different sensors, viewpoints, or weather conditions cause different domains. Models trained on domain A applied to tasks of domain B result in low performance. To overcome the domain shift, we propose an unsupervised pedestrian detection method that utilizes CycleGAN to establish an intermediate domain and transform a large gap domain-shift problem into two feature alignment subtasks with small gaps. The intermediate domain trained with labels from domain A, after two rounds of feature alignment using adversarial learning, can facilitate effective detection in domain B. To further enhance the training quality of intermediate domain models, Image Quality Assessment (IQA) is incorporated. The experimental results evaluated on Citypersons, KITTI, and BDD100K show that MR of 24.58%, 33.66%, 28.27%, and 28.25% were achieved in four cross-domain scenarios. Compared with typical pedestrian detection models, our proposed method can better overcome the domain-shift problem and achieve competitive results. |
| first_indexed | 2024-04-24T10:57:00Z |
| format | Article |
| id | doaj.art-05fd2b4da2bc4cbf8909473f133221e0 |
| institution | Directory Open Access Journal |
| issn | 2589-0042 |
| language | English |
| last_indexed | 2024-04-24T10:57:00Z |
| publishDate | 2024-04-01 |
| publisher | Elsevier |
| record_format | Article |
| series | iScience |
| spelling | doaj.art-05fd2b4da2bc4cbf8909473f133221e02024-04-12T04:45:43ZengElsevieriScience2589-00422024-04-01274109639Cross-domain pedestrian detection via feature alignment and image quality assessmentJun Yao0Zhilin Guo1JunJie Yu2Nan Yan3Qiong Wang4Wei Yu5The Engineering&Technical College of Chengdu University of Technology, Xiaoba Road, Leshan 614000, China; Corresponding authorThe Engineering&Technical College of Chengdu University of Technology, Xiaoba Road, Leshan 614000, ChinaThe Engineering&Technical College of Chengdu University of Technology, Xiaoba Road, Leshan 614000, ChinaThe Engineering&Technical College of Chengdu University of Technology, Xiaoba Road, Leshan 614000, ChinaThe Engineering&Technical College of Chengdu University of Technology, Xiaoba Road, Leshan 614000, ChinaThe Engineering&Technical College of Chengdu University of Technology, Xiaoba Road, Leshan 614000, China; China University of Mining and Technology, Daxue Road, Xuzhou 221116, China; Corresponding authorSummary: Datasets collected under different sensors, viewpoints, or weather conditions cause different domains. Models trained on domain A applied to tasks of domain B result in low performance. To overcome the domain shift, we propose an unsupervised pedestrian detection method that utilizes CycleGAN to establish an intermediate domain and transform a large gap domain-shift problem into two feature alignment subtasks with small gaps. The intermediate domain trained with labels from domain A, after two rounds of feature alignment using adversarial learning, can facilitate effective detection in domain B. To further enhance the training quality of intermediate domain models, Image Quality Assessment (IQA) is incorporated. The experimental results evaluated on Citypersons, KITTI, and BDD100K show that MR of 24.58%, 33.66%, 28.27%, and 28.25% were achieved in four cross-domain scenarios. Compared with typical pedestrian detection models, our proposed method can better overcome the domain-shift problem and achieve competitive results.http://www.sciencedirect.com/science/article/pii/S2589004224008617Computer scienceArtificial intelligenceEngineering |
| spellingShingle | Jun Yao Zhilin Guo JunJie Yu Nan Yan Qiong Wang Wei Yu Cross-domain pedestrian detection via feature alignment and image quality assessment iScience Computer science Artificial intelligence Engineering |
| title | Cross-domain pedestrian detection via feature alignment and image quality assessment |
| title_full | Cross-domain pedestrian detection via feature alignment and image quality assessment |
| title_fullStr | Cross-domain pedestrian detection via feature alignment and image quality assessment |
| title_full_unstemmed | Cross-domain pedestrian detection via feature alignment and image quality assessment |
| title_short | Cross-domain pedestrian detection via feature alignment and image quality assessment |
| title_sort | cross domain pedestrian detection via feature alignment and image quality assessment |
| topic | Computer science Artificial intelligence Engineering |
| url | http://www.sciencedirect.com/science/article/pii/S2589004224008617 |
| work_keys_str_mv | AT junyao crossdomainpedestriandetectionviafeaturealignmentandimagequalityassessment AT zhilinguo crossdomainpedestriandetectionviafeaturealignmentandimagequalityassessment AT junjieyu crossdomainpedestriandetectionviafeaturealignmentandimagequalityassessment AT nanyan crossdomainpedestriandetectionviafeaturealignmentandimagequalityassessment AT qiongwang crossdomainpedestriandetectionviafeaturealignmentandimagequalityassessment AT weiyu crossdomainpedestriandetectionviafeaturealignmentandimagequalityassessment |