提示
基于内容引导注意力的车道线检测网络
首发时间：2024-03-18
摘要：现代车道线检测方法在复杂的实际场景中取得了显著的性能。尽管注意力机制被广泛应用于许多计算机视觉任务，但在车道线检测任务中，如何更有效地利用注意力机制以提高准确性仍然是一个值得探索的领域。因此，本文提出了基于内容引导注意力的车道线检测网络（cganet），设计一种全新的内容引导注意力机制，旨在强调编码在特征中更有用的信息。为了实现多尺度特征的均衡融合，提出了一种均衡特征金字塔网络。此外，在损失函数中增加交叉熵损失，以提升模型精度。实验表明，对比现有最好的车道线检测方法，本文提出的方法在tusimple测试集上f1指标提升1.69%，acc指标提升1.27%。
关键词：
for information in english, please click here
content-guided attention-based lane detection network
abstract：modern lane detection methods have achieved significant performance in complex real-world scenarios. despite the widespread application of attention mechanisms in many computer vision tasks, how to more effectively utilize attention mechanisms to improve accuracy in lane detection tasks remains an area worth exploring. therefore, this paper proposed a feature-guided attention-based lane detection method (cganet), introducing a novel content-guided attention mechanism designed to emphasize information encoding more useful features. to achieve balanced fusion of multi-scale features, a balanced feature pyramid network was proposed. additionally, cross-entropy loss was incorporated into the loss function to enhance model accuracy. experiments show that, compared to the current state-of-the-art lane detection clrnet model, the proposed method in this paper achieves a 1.69% improvement in the f1 metric and a 1.27% improvement in the acc metric on the tusimple test set.
keywords：
基金：
论文图表：
引用
导出参考文献
no.****
动态公开评议
共计0人参与
勘误表
基于内容引导注意力的车道线检测网络
评论
全部评论