Dynamic Warping Network for Semantic Video Segmentation
A major challenge for semantic video segmentation is how to exploit the spatiotemporal information and produce consistent results for a video sequence. Many previous works utilize the precomputed optical flow to warp the feature maps across adjacent frames. However, the imprecise optical flow and th...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2021-01-01
|
| Series: | Complexity |
| Online Access: | http://dx.doi.org/10.1155/2021/6680509 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832560121364873216 |
|---|---|
| author | Jiangyun Li Yikai Zhao Xingjian He Xinxin Zhu Jing Liu |
| author_facet | Jiangyun Li Yikai Zhao Xingjian He Xinxin Zhu Jing Liu |
| author_sort | Jiangyun Li |
| collection | DOAJ |
| description | A major challenge for semantic video segmentation is how to exploit the spatiotemporal information and produce consistent results for a video sequence. Many previous works utilize the precomputed optical flow to warp the feature maps across adjacent frames. However, the imprecise optical flow and the warping operation without any learnable parameters may not achieve accurate feature warping and only bring a slight improvement. In this paper, we propose a novel framework named Dynamic Warping Network (DWNet) to adaptively warp the interframe features for improving the accuracy of warping-based models. Firstly, we design a flow refinement module (FRM) to optimize the precomputed optical flow. Then, we propose a flow-guided convolution (FG-Conv) to achieve the adaptive feature warping based on the refined optical flow. Furthermore, we introduce the temporal consistency loss including the feature consistency loss and prediction consistency loss to explicitly supervise the warped features instead of simple feature propagation and fusion, which guarantees the temporal consistency of video segmentation. Note that our DWNet adopts extra constraints to improve the temporal consistency in the training phase, while no additional calculation and postprocessing are required during inference. Extensive experiments show that our DWNet can achieve consistent improvement over various strong baselines and achieves state-of-the-art accuracy on the Cityscapes and CamVid benchmark datasets. |
| format | Article |
| id | doaj-art-80fa088562bd490896efaddb2eaac69b |
| institution | Kabale University |
| issn | 1076-2787 1099-0526 |
| language | English |
| publishDate | 2021-01-01 |
| publisher | Wiley |
| record_format | Article |
| series | Complexity |
| spelling | doaj-art-80fa088562bd490896efaddb2eaac69b2025-02-03T01:28:25ZengWileyComplexity1076-27871099-05262021-01-01202110.1155/2021/66805096680509Dynamic Warping Network for Semantic Video SegmentationJiangyun Li0Yikai Zhao1Xingjian He2Xinxin Zhu3Jing Liu4School of Automation & Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, ChinaSchool of Automation & Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, ChinaNational Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing 100083, ChinaNational Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing 100083, ChinaSchool of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100083, ChinaA major challenge for semantic video segmentation is how to exploit the spatiotemporal information and produce consistent results for a video sequence. Many previous works utilize the precomputed optical flow to warp the feature maps across adjacent frames. However, the imprecise optical flow and the warping operation without any learnable parameters may not achieve accurate feature warping and only bring a slight improvement. In this paper, we propose a novel framework named Dynamic Warping Network (DWNet) to adaptively warp the interframe features for improving the accuracy of warping-based models. Firstly, we design a flow refinement module (FRM) to optimize the precomputed optical flow. Then, we propose a flow-guided convolution (FG-Conv) to achieve the adaptive feature warping based on the refined optical flow. Furthermore, we introduce the temporal consistency loss including the feature consistency loss and prediction consistency loss to explicitly supervise the warped features instead of simple feature propagation and fusion, which guarantees the temporal consistency of video segmentation. Note that our DWNet adopts extra constraints to improve the temporal consistency in the training phase, while no additional calculation and postprocessing are required during inference. Extensive experiments show that our DWNet can achieve consistent improvement over various strong baselines and achieves state-of-the-art accuracy on the Cityscapes and CamVid benchmark datasets.http://dx.doi.org/10.1155/2021/6680509 |
| spellingShingle | Jiangyun Li Yikai Zhao Xingjian He Xinxin Zhu Jing Liu Dynamic Warping Network for Semantic Video Segmentation Complexity |
| title | Dynamic Warping Network for Semantic Video Segmentation |
| title_full | Dynamic Warping Network for Semantic Video Segmentation |
| title_fullStr | Dynamic Warping Network for Semantic Video Segmentation |
| title_full_unstemmed | Dynamic Warping Network for Semantic Video Segmentation |
| title_short | Dynamic Warping Network for Semantic Video Segmentation |
| title_sort | dynamic warping network for semantic video segmentation |
| url | http://dx.doi.org/10.1155/2021/6680509 |
| work_keys_str_mv | AT jiangyunli dynamicwarpingnetworkforsemanticvideosegmentation AT yikaizhao dynamicwarpingnetworkforsemanticvideosegmentation AT xingjianhe dynamicwarpingnetworkforsemanticvideosegmentation AT xinxinzhu dynamicwarpingnetworkforsemanticvideosegmentation AT jingliu dynamicwarpingnetworkforsemanticvideosegmentation |