A-WetDri
收藏资源简介:
A-WetDri是一个高分辨率的数据集,旨在提升自动驾驶系统在恶劣天气条件下的鲁棒性和泛化能力。该数据集包含模拟场景和真实世界驾驶视频,涵盖了雨天、雾天、夜间、雪天和晴朗天气等多种环境。数据集分为八种配置,包括模拟和真实世界场景、标注数据和原始图像,支持对象检测、域适应、模型训练和评估等研究。此外,README还提供了如何加载和使用数据集的示例代码。
A-WetDri is a high-resolution dataset designed to enhance the robustness and generalization capability of autonomous driving systems under adverse weather conditions. This dataset includes simulated scenarios and real-world driving videos, covering various environments such as rainy, foggy, nighttime, snowy, and clear weather conditions. The dataset is divided into eight configurations, including simulated and real-world scenarios, annotated data and raw images, which supports research in areas like object detection, domain adaptation, model training and evaluation. In addition, the README provides sample code demonstrating how to load and utilize the dataset.
A-WetDri 数据集概述
数据集基本信息
- 名称: A-WetDri (Adverse Weather Driving Dataset)
- 任务类别: 目标检测 (object-detection)
- 标签:
- 自动驾驶 (self_driving)
- 数据增强 (data_augmentation)
- 目标检测 (object_detection)
- 许可证: CC BY 4.0
- 数据集大小: 1,098,501,024.728 字节
- 下载大小: 2,890,471,847 字节
数据集结构
特征字段
file_name: 文件名 (string)image: 图像 (image)weather_condition: 天气条件 (string)bbox: 边界框 (sequence of float32)categories: 类别标签 (sequence of string)
数据配置
共有8种配置,分为以下两类:
实验性分割(含标注)
simulated_baseline: CARLA默认天气场景simulated_augmented: CARLA增强天气场景(雾、雨、夜)real_baseline: 真实世界晴朗天气场景real_augmented: 真实世界增强天气场景(雪、雾、雨、夜)
原始图像(按天气分组)
simulated_trainable_images: 合成训练图像simulated_evaluation_images: 合成评估图像real_trainable_images: 真实世界训练图像real_evaluation_images: 真实世界评估图像
数据统计
图像数量
| 框架 | 总图像数 | 训练/验证 | 测试 |
|---|---|---|---|
| 模拟数据 | 19,344 | 18,328 | 1,016 |
| 真实数据 | 23,046 | 22,156 | 1,250 |
分割统计(真实图像)
| 分割名称 | 样本数 | 大小(字节) |
|---|---|---|
| Real_Images_Evaluation_Set_Images_default | 250 | 8,597,689.0 |
| Real_Images_Evaluation_Set_Images_fog | 250 | 77,398,667.0 |
| Real_Images_Evaluation_Set_Images_night | 250 | 79,622,113.0 |
| Real_Images_Evaluation_Set_Images_rain | 250 | 89,146,938.0 |
| Real_Images_Evaluation_Set_Images_snow | 250 | 98,299,404.0 |
| Real_Images_Trainable_Set_Images_default | 7,790 | 270,086,905.32 |
| Real_Images_Trainable_Set_Images_fog | 2,981 | 69,475,588.595 |
| Real_Images_Trainable_Set_Images_night | 4,137 | 102,642,932.143 |
| Real_Images_Trainable_Set_Images_rain | 3,449 | 139,785,980.284 |
| Real_Images_Trainable_Set_Images_snow | 3,799 | 163,444,807.386 |
对象类别
- 汽车 (Car)
- 卡车 (Truck)
- 巴士 (Bus)
- 自行车 (Bicycle)
- 摩托车 (Motorcycle)
- 行人 (Walker)
- 交通标志 (TrafficSigns)
- 交通信号灯 (TrafficLight)
使用示例
python from datasets import load_dataset
加载真实增强数据集
dataset = load_dataset( "unai-gurbindo/A-WetDri", name="real_augmented", split="train" )
访问图像和边界框
image = dataset[0]["image"] bboxes = dataset[0]["bbox"]
数据来源
- 模拟数据: CARLA
- 真实数据: BDD100K 和 ACDC




