基于跨任务双向特征交互的交通场景感知算法

林鹏志,钟铭恩,范康,谭佳威,林志强

Traffic scene perception algorithm based on cross-task bidirectional feature interaction

Pengzhi LIN,Ming’en ZHONG,Kang FAN,Jiawei TAN,Zhiqiang LIN

表 5 SDFormer++与单任务深度估计算法的性能对比

Tab.5 Performance comparison of SDFormer++ and single-task depth estimation algorithms

方法 骨干网络 RMSE ARE Np/106 GFLOPs Manydepth2 HRNet 16 5.827 0.097 123.1 246.4 DepthFormer Swin-B 4.326 0.127 151.3 282.0 PixelFormer Swin-B 4.258 0.115 146.1 346.4 SDFormer++ Swin-B 4.453 0.130 129.4 272.5