DDD17公开数据集

本项目基于DDD17公开数据获取DDD17-Semantic数据集，进行感知相关算法的验证，DDD17-Semantic数据集是首个同步包含事件帧图像和APS图像的行车记录数据集，主要面向事件相机和语义分割研究。该数据集基于DDD17数据集，筛选出六个对比度适中的序列。首先在Cityscapes数据集上训练灰度图像分割模型，达到83%的mIoU，然后使用该模型对DDD17的APS帧图像进行预测，生成事件帧图像的语义标签。最终数据集大小为1.1G，包含9440张图片，涵盖6个类别，适用于事件相机语义分割任务的研究与开发。This project develops the DDD17-Semantic dataset based on the publicly available DDD17 dataset for validating perception-related algorithms. The DDD17-Semantic dataset is the first driving recording dataset that simultaneously includes event frame images and APS images, primarily targeting event camera and semantic segmentation research. Derived from the original DDD17 dataset, this work selects six sequences with moderate contrast levels. First, a grayscale image segmentation model is trained on the Cityscapes dataset, achieving an mIoU of 83%. Subsequently, this model is used to predict the APS frame images of DDD17, generating semantic labels for the event frame images. The final dataset has a size of 1.1 GB, contains 9,440 images across 6 categories, and is suitable for research and development of event camera semantic segmentation tasks.

搜集汇总

背景与挑战

背景概述

DDD17公开数据集基于DDD17数据生成DDD17-Semantic数据集，这是首个同步包含事件帧和APS图像的行车记录数据集，专注于事件相机和语义分割研究。该数据集包含9440张图片，大小约1.1G，涵盖6个类别，适用于自动驾驶相关算法的验证与开发。