Carla场景仿真数据集

本数据集依托自动驾驶仿真平台Carla，采用两种采集方案。方案一分别采集了车辆在乡村、城镇、交叉路口、高速路四种场景下的传感器信息，重点关注RGB相机、深度相机和激光雷达三种传感器，收集时间同步的RGB图像、深度图像、点云三种感知数据，数据量为294M。通过该数据集可以验证自动驾驶深度感知算法在不同场景下的性能，评估相关深度感知模型的性能差异。方案二采集了1000个左右连续的动态场景，涵盖多种天气条件，光线条件，不同交通参与者以及不同场所和道路类型等。

This dataset is built upon the autonomous driving simulation platform Carla, with two data collection schemes adopted. Scheme 1 collects sensor information of the vehicle in four scenarios: rural, urban, intersection, and highway. It focuses on three types of sensors, namely RGB cameras, depth cameras, and LiDAR, and gathers three kinds of time-synchronized perception data: RGB images, depth maps, and point clouds, with a total data volume of 294 MB. This dataset can be used to verify the performance of autonomous driving depth perception algorithms across different scenarios, and evaluate the performance differences of relevant depth perception models. Scheme 2 collects approximately 1,000 continuous dynamic scenarios, covering various weather conditions, lighting conditions, different traffic participants, as well as different locations and road types, etc.

搜集汇总

背景与挑战

背景概述

Carla场景仿真数据集是一个用于自动驾驶算法验证的仿真数据集，包含两种采集方案：方案一提供了乡村、城镇等四种场景下的RGB图像、深度图像和点云同步数据，专注于深度感知算法评估；方案二包含约1000个连续动态场景，覆盖多样化的天气、光照和交通条件。数据集总规模为353.54MB，由西北工业大学发布，支持自动驾驶模型在不同环境下的性能测试。