Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 5, 740(2025)
Scene recognition based on deep metric learning and semantic segmentation
Fig. 1. Overall network architecture of the improved Deeplabv3+ model
Fig. 8. Comparison of verification accuracy between DML module and baseline module
Fig. 9. Influence of different loss functions on verification accuracy
Fig. 10. Comparison of segmentation results between the module in this paper and the baseline module
Table 1. Comparison of the method in this paper with state-of-the-art methods on ADE20K validation set
View tableView in ArticleTable 1. Comparison of the method in this paper with state-of-the-art methods on ADE20K validation set
算法 mIoU/% Pixel-Acc/% PSPNet［ 19 ］43.29 81.39 CFNNet［ 20 ］44.89 82.40 CCNet［ 21 ］45.22 79.83 OCR［ 22 ］45.28 82.52 GFFNet［ 23 ］45.33 82.01 GSCNet［ 24 ］45.89 82.01 CTNet［ 25 ］45.94 82.28 本文方法 47.60 82.60
Table 2. Comparison of the method in this paper with state-of-the-art methods on the Cityscapes dataset
View tableView in ArticleTable 2. Comparison of the method in this paper with state-of-the-art methods on the Cityscapes dataset
算法 mIoU% Backbone DeepLabv3 plus+BiFPN+Shuffle Attention［ 26 ］78.69 ResNet50 Mask DINO［ 27 ］80.0 ResNet50 HRNetV2［ 28 ］81.1 HRNetV2-w48 CFANet［ 29 ］81.34 ResNet50 CCNet［ 21 ］81.4 ResNet101 GFFNet［ 23 ］82.3 ResNet101 GSCNet［ 24 ］82.6 ResNet101 CTNet［ 25 ］82.8 HRNetV2-w48 OCR［ 22 ］83.0 ResNet101 本文方法 83.1 ResNet101
Table 3. Ablation experiments of DML module on ADE20K validation set
View tableView in ArticleTable 3. Ablation experiments of DML module on ADE20K validation set
算法 mIoU/% Pixel-Acc/% 基线模型 46.5 81.01 +DML 47.6 82.6
Table 4. Experimental ablation of loss function on ADE20K validation set
View tableView in ArticleTable 4. Experimental ablation of loss function on ADE20K validation set
算法 mIoU/% 基线模型 45.23 +焦点损失 46.54 +焦点损失+对比损失 47.6
Get Citation
Copy Citation Text
Xuan JIA, Ye ZHANG, Xuling CHANG, Jianbo SUN. Scene recognition based on deep metric learning and semantic segmentation[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(5): 740
Category:
Received: Sep. 20, 2024
Accepted: --
Published Online: Jun. 18, 2025
The Author Email: Ye ZHANG (yolanda@spirits.ai)
