CURE-TSD (CURE Traffic Sign Detection)

CURE-TSD 数据集中的视频序列分为两类：真实数据和非真实数据。真实数据对应于从真实世界获取的序列的处理版本。虚幻数据对应于在虚拟环境中生成的合成序列。有 49 个真实序列和 49 个不真实序列，不包括任何特定挑战。我们将序列分成 70% 和 %30 的分割。因此，我们有 34 个训练视频和 15 个测试视频，包括真实和虚幻序列，没有挑战。每个视频序列有 300 帧。共处理了 49 个无挑战的真实视频序列，采用 12 种不同类型的效果和 5 个不同的挑战级别，产生 2,989 (49125+49) 个视频序列。此外，有 49 个合成视频序列处理了 11 种不同类型的效果和 5 个不同的挑战级别，从而产生了 2,744 (49115+49) 个视频序列。总共有 5,733 个视频序列，其中包括大约 172 万帧。

The video sequences in the CURE-TSD dataset are categorized into two classes: real data and non-real data. Real data refers to the processed versions of sequences captured from the real world, while non-real data corresponds to synthetic sequences generated in virtual environments. Initially, there are 49 real sequences and 49 non-real sequences without any specific challenges. We split these sequences into a 70% training set and 30% test set, resulting in 34 training videos and 15 test videos, which include both real and non-real sequences with no additional challenges. Each video sequence contains 300 frames. A total of 49 unchallenged real video sequences were processed with 12 distinct types of visual effects and 5 different challenge levels, yielding 2,989 video sequences (calculated as 49×12×5 + 49). Additionally, 49 synthetic video sequences were processed with 11 distinct types of visual effects and 5 different challenge levels, producing 2,744 video sequences (calculated as 49×11×5 + 49). Overall, the dataset comprises 5,733 video sequences in total, encompassing approximately 1.72 million frames.

搜集汇总

背景与挑战

背景概述

CURE-TSD是一个专注于挑战性条件下交通标志检测的数据集，包含真实和合成视频序列，总共有5,733个视频约172万帧，用于训练和测试模型。数据集由佐治亚理工学院于2017年发布，旨在评估交通标志检测在复杂场景中的性能变化和光谱特性。