D²-City

D ²-City是一个大型驾驶视频数据集，提供10,000多个行车记录仪视频记录在720p HD或1080p FHD。大约1000个视频在所有道路对象的每个帧中都带有检测和跟踪注释，包括边界框以及汽车，货车，公共汽车，卡车，行人，摩托车，自行车，敞开式和封闭式三轮车，叉车的跟踪id，以及大型和小型街区。视频的一些剩余部分带有在关键帧中注释的道路对象。

D²-City is a large-scale driving video dataset containing over 10,000 dashcam videos recorded in 720p HD or 1080p FHD resolution. Approximately 1,000 of these videos are fully annotated with detection and tracking information for every frame across all road objects, including bounding boxes and tracking IDs for cars, vans, buses, trucks, pedestrians, motorcycles, bicycles, open-top and closed tricycles, and forklifts, as well as large and small urban blocks. For the remaining segments of the dataset, road objects are annotated only in key frames.

搜集汇总

背景与挑战

背景概述

D²-City是一个大型驾驶视频数据集，包含超过10,000个行车记录仪视频，分辨率高达1080p FHD，主要用于自动驾驶和计算机视觉研究。其特点在于约1000个视频提供了逐帧的检测和跟踪注释，涵盖多种道路对象如汽车、行人和自行车，其余视频则在关键帧中注释。该数据集由南加州大学·滴滴实验室于2021年发布，遵循CC BY 4.0许可，数据量达131.4GB，支持对象检测和跟踪任务。