The DADA-2000 dataset is a large-scale driving attention dataset for driving accidents within the driving field of view. The dataset contains 2,000 driving accident sequences, totaling 658,476 frames, which are divided into 54 categories according to whether the vehicle is involved and the type of participants at the time of the accident. The dataset covers different weather, lighting, and road environments. In addition, the dataset records the attention data of each frame (including the gaze spectrum and scanning path), and manually annotates the real abnormal coordinates of each sequence.