DADA-2000 驾驶注意力数据集

DADA-2000 数据集是面向驾驶视野内行车事故的大型行车驾驶注意力数据集。该数据集共 2,000 个驾驶事故序列，共计 658,476 帧，根据自车参与与否及事故发生时的参与者类型分为 54 类。该数据集涵盖不同天气，光照，道路环境。此外，该数据集记录了每一帧的注意力数据（包括注视图谱与扫视路径），并且手工标注了每个序列的真实异常坐标。

The DADA-2000 dataset is a large-scale driving attention dataset targeting traffic accidents within the driver's field of view. It contains 2,000 driving accident sequences, totaling 658,476 frames, and is classified into 54 categories based on whether the ego vehicle is involved and the types of participants during the accident. This dataset covers diverse weather conditions, lighting scenarios and road environments. In addition, it records attention data for each frame, including fixation maps and scan paths, and manually annotates the ground-truth abnormal coordinates for each sequence.

搜集汇总

背景与挑战

背景概述

DADA-2000 是一个专注于驾驶事故场景的大型注意力数据集，包含2,000个事故序列和超过65万帧图像，覆盖多种环境条件并精细分类为54类。该数据集独特之处在于提供了每帧的注意力数据（如注视图谱和扫视路径）以及手工标注的真实异常坐标，旨在支持自动驾驶和语义分割领域的研究，特别是事故预测和注意力机制分析。