Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 7, 990(2024)
Semantic segmentation algorithm for foggy cityscapes images by fusing self-supervised contrastive learning
Fig. 4. Variation curves of training accuracy of different networks
Fig. 5. Visualization results of different networks in misty weather
Fig. 6. Comparison of segmentation accuracy of different categories
Fig. 8. Trend diagram of optimal values for different fog concentration
Fig. 9. Visualization results of different algorithms in moderate fog weather
Fig. 10. Visualization results of different algorithms in dense fog weather
Table 1. Results of ablation experiments
View tableView in ArticleTable 1. Results of ablation experiments
Baseline DA-ASPP 自监督边缘提取模块 LSCFNet MIoU/% √ 70.53 √ 71.60 √ √ 72.97 √ √ √ 74.35
Table 2. Results of comparison experiments
View tableView in ArticleTable 2. Results of comparison experiments
Model MIoU/% MPA/% PA/% Params/106 FLOPs/G Time/s BiseNetV2 61.39 71.77 92.02 12.80 13.67 37.54 ICNet 59.98 68.25 92.27 28.30 36.99 97.08 PSPNet 66.56 74.45 93.17 67.76 270.53 583.14 DeepLabV3+ 70.53 79.60 94.83 5.23 16.95 64.05 SCFNet 74.35 83.59 95.85 2.88 15.10 62.15
Table 3. Street segmentation results for different concentrations of foggy days
View tableView in ArticleTable 3. Street segmentation results for different concentrations of foggy days
Foggy dataset Model MIoU/% MPA/% PA/% 薄雾 DeepLabV3+SCFNet 70.5374.35 79.6083.59 94.8395.85 中雾 DeepLabV3+SCFNet 70.7272.29 79.3781.73 94.7795.35 浓雾 DeepLabV3+SCFNet 69.6672.41 78.9482.07 94.4595.37
Get Citation
Copy Citation Text
Liwei LIU, Rui WANG, Xutao MENG. Semantic segmentation algorithm for foggy cityscapes images by fusing self-supervised contrastive learning[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(7): 990
Category: Research Articles
Received: May. 29, 2023
Accepted: --
Published Online: Jul. 23, 2024
The Author Email: Liwei LIU (liuliwei@ccut.edu.cn)
