Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 9, 1321(2025)
River edge detection based on improved EDTER algorithm
Fig. 8. Examples of training samples. （a）~（d） Original images； （e）~（h） Corresponding label maps.
Fig. 9. Visual comparison of model performance on CRRE dataset. （a） Original images； （b） BDCN； （c） RCF； （d） EDTER； （e） Ours.
Table 1. Comparison of different backbone networks on ImageNet-1K classification
View tableView in ArticleTable 1. Comparison of different backbone networks on ImageNet-1K classification
模型 图片尺寸/像素 参数量/M FLOPs/G 吞吐率/（张·s-1） 准确率/% ViT-B/16［ 13 ］384×384 86 55.4 85.9 77.9 ViT-L/16［ 13 ］384×384 307 190.7 27.3 76.5 Swin-T［ 13 ］224×224 29 4.5 755.2 81.3 Swin-S［ 14 ］224×224 50 8.7 436.9 83.0 Swin-B［ 14 ］224×224 88 15.4 278.1 83.5 Swin-B［ 14 ］224×224 88 47.0 84.7 84.5
Table 2. Ablation experiments on BSDS500 dataset
View tableView in ArticleTable 2. Ablation experiments on BSDS500 dataset
No. Swin-T Res-MLP Lite MSD ODS OIS AP 1 0.824 0.841 0.880 2 √ 0.827 0.845 0.881 3 √ 0.826 0.843 0.878 4 √ 0.825 0.842 0.882 5 √ √ √ 0.835 0.848 0.884
Table 3. Comparison of model performance on BSDS500 dataset
View tableView in ArticleTable 3. Comparison of model performance on BSDS500 dataset
模型 ODS OIS AP BDCN［ 24 ］0.806 0.826 0.890 HED［ 25 ］0.788 0.808 0.840 EDTER［ 15 ］0.824 0.841 0.880 DeepContour［ 27 ］0.757 0.776 0.800 DeepEdge［ 28 ］0.753 0.772 0.807 PiDiNet［ 29 ］0.789 0.803 - RCF［ 26 ］0.811 0.830 - Ours 0.835 0.848 0.884
Table 4. Comparison of model performance on CRRE dataset
View tableView in ArticleTable 4. Comparison of model performance on CRRE dataset
模型 ODS OIS AP BDCN［ 24 ］0.821 0.835 0.845 HED［ 25 ］0.812 0.824 0.834 EDTER［ 15 ］0.825 0.831 0.839 RCF［ 26 ］0.817 0.833 - Ours 0.841 0.849 0.871
Get Citation
Copy Citation Text
Yudong GAO, Lingxiao HUANG, Xinbo YAO, Yuanru ZHAO. River edge detection based on improved EDTER algorithm[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(9): 1321
Category:
Received: May. 21, 2025
Accepted: --
Published Online: Sep. 25, 2025
The Author Email: Lingxiao HUANG (huanglx@nxu.edu.cn)
