[车道线检测]Keep your Eyes on the Lane: Real-time Attention-guided Lane Detection

论文地址：Keep your Eyes on the Lane: Real-time Attention-guided Lane Detection
解决real-time efficiency. LaneATT是一个anchor based的车道线检测模型。基于全局信息来推理车道线信息，以此来解决车道线被占用，缺少markers等。

Introduction

1. 车道线检测的挑战：extreme light and weather, lane markings occluded by objects, data imbalance, real-time.
2. 利用anchor-based feature pooling 实现高准度和实时性

How

LaneATT是单阶段anchor-based模型。从前试图提取特征图，然后利用anchor-based feature pooling提取每个anchor的特征。这些特征和global infromation结合利用attention机制（可以解决occlusion），在经过全连接生成最后的输出车道线。

1. anchor-based feature pooling。anchor的表示: origin point $O=(x_{ori},y_{ori})$，处在图像边缘； direction$\theta$.
2. global information attention