Dataset for multi-camera depth reconstruction

Abstract

This dataset is composed from three different scenes with a varying number of takes per scene. The images have been generated using CARLA, a framework for autonomeous driving. In the dataset, an autonomeous car is exploring an urban area under different lighting and weather conditions.

In each take, the car is equipped with three front-facing cameras capturing the scenery. Besides pure RGB image data, the dataset features ground-truth depth information encoded as described here: https://carla.readthedocs.io/en/latest/ref_sensors/#depth-camera

The camera parameters and capture setup is as follows:

Field of view: 110° Resolution: 1920 x 1080 Number of takes: Scene 1: 41 Scene 2: 23 Scene 3: 20 Camera baseline Scene 1: 0.280833m (2.17m/12) Scene 2 and 3: 0.1m