Weakly But Deeply Supervised Occlusion-Reasoned Parametric Road Layouts

We propose an end-to-end network that takes a single perspective RGB image of a complex road scene as input, to produce occlusion-reasoned layouts in perspective space as well as a parametric bird's-eye-view (BEV) space. In contrast to prior works that require dense supervision such as semantic labels in perspective view, our method only requires human annotations for parametric attributes that are cheaper and less ambiguous to obtain. To solve this challenging task, our design is comprised of modules that incorporate inductive biases to learn occlusion-reasoning, geometric transformation and semantic abstraction, where each module may be supervised by appropriately transforming the parametric annotations. We demonstrate how our design choices and proposed deep supervision help achieve meaningful representations and accurate predictions. We validate our approach on two public datasets, KITTI and NuScenes, to achieve state-of-the-art results with considerably less human supervision.

1. Introduction

Understanding road layout from images is essential for real-world applications such as autonomous driving or path planning [5], [8], [13], [31], where, besides the usual perspective space outputs, top-view representations of geometry and semantics have been popular. Non-parametric representations such as pixel-level semantics [31] generally require labor-intensive and potentially ambiguous supervision in the top-view, for example, when dealing with occluded regions. On the other hand, parametric representations for top-view layouts are desirable for their interpretability, which is beneficial for higher-level reasoning and decision-making in downstream applications.