基于内容引导注意力的车道线检测网络

刘登峰,郭文静,陈世海

Content-guided attention-based lane detection network

Dengfeng LIU,Wenjing GUO,Shihai CHEN

表 2 CGANet在CULane上的实验结果

Tab.2 CGANet’s experimental results on CULane dataset

方法 基线网络 F1/% mF1/% F1/% Ncross FPS/帧 Flops/109 Normal Crowd Dazzle Shadow No line Arrow Curve Ningt SCNN[8] VGG16 71.60 38.34 90.60 69.70 58.50 66.90 43.40 84.10 64.40 66.10 1990 7.5 328.4 UFLD[2] ResNet18 68.40 38.94 85.90 63.60 57.00 69.60 40.60 79.40 65.20 66.70 2037 282.0 8.4 ResNet34 72.30 38.96 87.70 66.00 58.40 68.80 40.20 81.00 57.90 62.10 1473 170.0 16.9 LaneATT[15] ResNet18 74.50 47.35 90.71 69.71 61.82 64.03 47.13 86.82 64.75 66.58 1020 153.0 9.3 ResNet34 74.00 47.57 91.14 72.03 62.47 74.15 47.39 87.38 64.75 68.72 1330 130.0 18.0 ResNet122 74.40 48.48 90.74 69.74 65.47 72.31 48.46 85.29 68.72 68.81 1264 21.0 70.5 CondLaneNet[14] ResNet18 75.13 48.84 91.87 74.87 66.72 76.01 50.39 88.37 72.40 71.23 1364 175.0 10.2 ResNet34 76.68 49.11 92.38 74.14 67.17 75.93 49.85 88.89 72.88 71.92 1387 128.0 19.6 ResNet101 77.02 50.83 92.47 74.14 66.93 76.91 52.13 89.16 72.21 72.80 1201 49.0 44.8 CLRerNet[3] ResNet18 76.12 52.11 92.60 75.92 70.23 77.33 52.34 88.57 72.68 73.25 1458 119.0 13.2 ResNet34 77.27 52.45 92.53 75.96 70.45 78.92 52.98 89.23 72.81 73.56 1334 104.0 24.5 ResNet101 78.80 52.68 92.80 76.12 69.84 78.95 53.65 89.69 73.45 73.37 1289 50.0 41.2 CLRNet[7] ResNet18 78.14 51.92 92.69 75.06 69.70 75.39 51.96 89.25 68.09 73.22 1520 119.0 12.9 ResNet34 78.74 51.14 92.49 75.33 70.57 75.92 52.01 89.59 72.77 73.02 1448 103.0 22.6 ResNet101 79.48 51.55 92.85 75.78 68.49 78.33 52.50 88.79 72.57 73.51 1456 46.0 40.5 CGANet (本研究方法) ResNet18 79.58 52.62 92.89 76.03 69.53 76.60 49.73 88.57 72.37 73.13 1321 120.0 13.7 ResNet34 79.73 52.31 92.87 75.86 70.57 76.88 50.03 89.79 73.23 73.74 1216 112.0 30.6 ResNet101 80.13 52.88 92.54 76.78 68.49 79.51 50.58 87.62 73.68 73.36 1262 57.0 42.7