尽管基于激光雷达的三维物体检测方法表现出优异的性能,但其传感器高成本和对天气的敏感性限制了广泛的应用。近年来，基于双目视觉的方法具有低成本以及较为满意的精度，从而受到越来越多的关注。尽管如此，基于双目的和基于激光雷达的三维目标检测方法之间仍存在巨大的性能差距。在该论文中，我们提出了StereoDistill，其核心的想法是在响应层面，让双目模型获取来自激光雷达模型有利的知识。在KIITI数据集上，与现有的双目方法对比，StereoDistill取得了最前沿的结果。
Although LiDAR-based 3D object detection methods have shown excellent performance, the high cost and sensitivity to weather of LiDAR sensors limit its wide application. In recent years, methods based on stereo images have attracted increasing attention due to their good balance of low cost and accuracy. Nevertheless, there is still a huge performance gap between these two methods. In this paper, we propose StereoDistill, whose key goal is to make the stereo model acquire the beneficial knowledge from LiDAR model in the response level. On the KIITI dataset, StereoDistill achieves state-of-the-art results compared to existing stereo-based approaches.
