HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking
Siamese network-based trackers consider tracking as features cross-correlation between the target template and the search region. Therefore, feature representation plays an important role for constructing a high-performance tracker. However, all existing Siamese networks extract the deep but low-res...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-08-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/20/17/4807 |
| _version_ | 1797555545362661376 |
|---|---|
| author | Dawei Zhang Zhonglong Zheng Tianxiang Wang Yiran He |
| author_facet | Dawei Zhang Zhonglong Zheng Tianxiang Wang Yiran He |
| author_sort | Dawei Zhang |
| collection | DOAJ |
| description | Siamese network-based trackers consider tracking as features cross-correlation between the target template and the search region. Therefore, feature representation plays an important role for constructing a high-performance tracker. However, all existing Siamese networks extract the deep but low-resolution features of the entire patch, which is not robust enough to estimate the target bounding box accurately. In this work, to address this issue, we propose a novel high-resolution Siamese network, which connects the high-to-low resolution convolution streams in parallel as well as repeatedly exchanges the information across resolutions to maintain high-resolution representations. The resulting representation is semantically richer and spatially more precise by a simple yet effective multi-scale feature fusion strategy. Moreover, we exploit attention mechanisms to learn object-aware masks for adaptive feature refinement, and use deformable convolution to handle complex geometric transformations. This makes the target more discriminative against distractors and background. Without bells and whistles, extensive experiments on popular tracking benchmarks containing OTB100, UAV123, VOT2018 and LaSOT demonstrate that the proposed tracker achieves state-of-the-art performance and runs in real time, confirming its efficiency and effectiveness. |
| first_indexed | 2024-03-10T16:49:02Z |
| format | Article |
| id | doaj.art-ad116e2aeb4f43e8a16e9ba8aa1ea0c6 |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-10T16:49:02Z |
| publishDate | 2020-08-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-ad116e2aeb4f43e8a16e9ba8aa1ea0c62023-11-20T11:23:15ZengMDPI AGSensors1424-82202020-08-012017480710.3390/s20174807HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object TrackingDawei Zhang0Zhonglong Zheng1Tianxiang Wang2Yiran He3Department of Computer Science, College of Mathematics and Computer Science, Zhejiang Normal University, No 688, Yingbin Road, Jinhua 321004, ChinaDepartment of Computer Science, College of Mathematics and Computer Science, Zhejiang Normal University, No 688, Yingbin Road, Jinhua 321004, ChinaDepartment of Computer Science, College of Mathematics and Computer Science, Zhejiang Normal University, No 688, Yingbin Road, Jinhua 321004, ChinaDepartment of Computer Science, College of Mathematics and Computer Science, Zhejiang Normal University, No 688, Yingbin Road, Jinhua 321004, ChinaSiamese network-based trackers consider tracking as features cross-correlation between the target template and the search region. Therefore, feature representation plays an important role for constructing a high-performance tracker. However, all existing Siamese networks extract the deep but low-resolution features of the entire patch, which is not robust enough to estimate the target bounding box accurately. In this work, to address this issue, we propose a novel high-resolution Siamese network, which connects the high-to-low resolution convolution streams in parallel as well as repeatedly exchanges the information across resolutions to maintain high-resolution representations. The resulting representation is semantically richer and spatially more precise by a simple yet effective multi-scale feature fusion strategy. Moreover, we exploit attention mechanisms to learn object-aware masks for adaptive feature refinement, and use deformable convolution to handle complex geometric transformations. This makes the target more discriminative against distractors and background. Without bells and whistles, extensive experiments on popular tracking benchmarks containing OTB100, UAV123, VOT2018 and LaSOT demonstrate that the proposed tracker achieves state-of-the-art performance and runs in real time, confirming its efficiency and effectiveness.https://www.mdpi.com/1424-8220/20/17/4807Siamese networkhigh-resolution representationmulti-scale fusionvisual trackingattention mechanismsdeformable convolution |
| spellingShingle | Dawei Zhang Zhonglong Zheng Tianxiang Wang Yiran He HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking Sensors Siamese network high-resolution representation multi-scale fusion visual tracking attention mechanisms deformable convolution |
| title | HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking |
| title_full | HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking |
| title_fullStr | HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking |
| title_full_unstemmed | HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking |
| title_short | HROM: Learning High-Resolution Representation and Object-Aware Masks for Visual Object Tracking |
| title_sort | hrom learning high resolution representation and object aware masks for visual object tracking |
| topic | Siamese network high-resolution representation multi-scale fusion visual tracking attention mechanisms deformable convolution |
| url | https://www.mdpi.com/1424-8220/20/17/4807 |
| work_keys_str_mv | AT daweizhang hromlearninghighresolutionrepresentationandobjectawaremasksforvisualobjecttracking AT zhonglongzheng hromlearninghighresolutionrepresentationandobjectawaremasksforvisualobjecttracking AT tianxiangwang hromlearninghighresolutionrepresentationandobjectawaremasksforvisualobjecttracking AT yiranhe hromlearninghighresolutionrepresentationandobjectawaremasksforvisualobjecttracking |