MGNet: Monocular Geometric Scene Understanding for Autonomous Driving

TL;DR

MGNet is a real-time multi-task neural network that combines panoptic segmentation and monocular depth estimation to produce dense 3D scene understanding for autonomous driving from single images.

Contribution

It is the first to unify panoptic segmentation and self-supervised monocular depth estimation in a single, low-latency model for autonomous driving.

Findings

Achieves competitive performance on Cityscapes and KITTI benchmarks.

Operates in real-time on consumer-grade GPUs.

Produces dense 3D point clouds with semantic labels from single images.

Abstract

We introduce MGNet, a multi-task framework for monocular geometric scene understanding. We define monocular geometric scene understanding as the combination of two known tasks: Panoptic segmentation and self-supervised monocular depth estimation. Panoptic segmentation captures the full scene not only semantically, but also on an instance basis. Self-supervised monocular depth estimation uses geometric constraints derived from the camera measurement model in order to measure depth from monocular video sequences only. To the best of our knowledge, we are the first to propose the combination of these two tasks in one single model. Our model is designed with focus on low latency to provide fast inference in real-time on a single consumer-grade GPU. During deployment, our model produces dense 3D point clouds with instance aware semantic labels from single high-resolution camera images. We evaluate our model on two popular autonomous driving benchmarks, i.e., Cityscapes and KITTI, and show competitive performance among other real-time capable methods. Source code is available at https://github.com/markusschoen/MGNet.