基于轻量级Transformer的城市路网提取方法
|
冯志成,杨杰,陈智超
|
Urban road network extraction method based on lightweight Transformer
|
Zhicheng FENG,Jie YANG,Zhichao CHEN
|
|
表 3 RoadViT和主流模型在不同数据集上的对比 |
Tab.3 Comparison of RoadViT and mainstream models on different datasets |
|
模型 | P/106 | FLOPs/109 | RIoU/% | CHN6-CUG 数据集 | DeepGlobe数据集 | PSPNet[16](ResNet18[11]) | 12.92 | 67.51 | 57.1 | 57.7 | DeepLab V3[15](ResNet18) | 13.60 | 85.97 | 57.6 | 58.6 | PSPNet(MobileNet V2[9]) | 2.65 | 10.72 | 55.3 | 54.5 | DeepLab V3(MobileNet V2) | 3.23 | 22.61 | 53.6 | 55.4 | LRASPP[10] | 3.22 | 1.98 | 51.1 | 51.1 | BiseNet V2[19] | 3.62 | 12.80 | 56.4 | 51.8 | STDC[18] | 14.23 | 23.51 | 60.7 | 54.6 | DDRNet[17] | 20.15 | 17.87 | 61.0 | 54.8 | PIDNet[20] | 7.62 | 5.89 | 60.0 | 52.6 | RoadViT | 1.25 | 1.18 | 57.0 | 52.3 | RoadViT-m | 2.35 | 3.02 | 58.7 | 53.7 | RoadViT-l | 5.97 | 6.01 | 59.7 | 54.3 | D-LinkNet[22] | — | — | 55.7 | — | HsgNet[22] | — | — | 57.7 | — |
|
|
|