PedSynth

PedSynth是由巴塞罗那自治大学计算机科学与CVC开发的合成数据集，专注于行人意图预测。该数据集包含947个视频片段，总计约398,000帧，每帧都标有行人是否将穿越的标签。数据集通过ARCANE框架生成，该框架允许程序化定义交通场景中的行人行为。PedSynth涵盖了多种天气和光照条件下的400个不同地点，旨在为自动驾驶系统提供更准确的行人意图预测模型。PedSynth is a synthetic dataset developed by the Department of Computer Science and the Computer Vision Center (CVC) at Universitat Autònoma de Barcelona, focusing on pedestrian intent prediction. It contains 947 video clips, totaling approximately 398,000 frames, with each frame annotated with a label indicating whether a pedestrian will cross the road. The dataset is generated via the ARCANE framework, which allows for programmatic definition of pedestrian behaviors in traffic scenarios. PedSynth covers 400 distinct locations under various weather and lighting conditions, aiming to provide more accurate pedestrian intent prediction models for autonomous driving systems.

搜集汇总

背景与挑战

背景概述

PedSynth是一个由巴塞罗那自治大学与CVC开发的合成数据集，专注于行人意图预测，包含947个视频片段（约398,000帧），每帧标注了行人是否将要穿越马路。该数据集通过ARCANE框架生成，覆盖400个不同地点及多种天气和光照条件，旨在为自动驾驶系统提供更准确的行人意图预测模型。