IDDAW_OFFICIAL
收藏资源简介:
IDD-AW(India Driving Dataset – Adverse Weather)是一个针对自动驾驶在恶劣天气条件下的语义分割基准数据集。该数据集基于印度驾驶场景,专门涵盖雨、雾、低光照和雪等恶劣天气条件,并提供了配对的RGB和近红外(NIR)图像,以及密集的Level-3语义标签(共26个类别)。数据集的训练集包含143个行车记录(drives),共3430帧;验证集包含18个行车记录,共475帧。每个帧对应三个已配准的PNG图像:RGB图像、NIR图像和Level-3语义标签图。标签图为单通道,像素值对应类别ID(0-25)。目录结构划分为train/和val/,每个子集下包含rgb/、nir/和gt_labels/三个子目录,以driveId和帧索引组织文件。该数据集适用于自动驾驶场景下的安全语义分割任务,特别是在恶劣天气条件下评估和提升分割模型的鲁棒性。
IDD-AW (India Driving Dataset – Adverse Weather) is a semantic segmentation benchmark dataset for autonomous driving under adverse weather conditions. Based on Indian driving scenarios, it specifically covers adverse weather conditions such as rain, fog, low light, and snow, and provides paired RGB and near-infrared (NIR) images along with dense Level-3 semantic labels (26 categories). The training set contains 143 drives (3430 frames), and the validation set contains 18 drives (475 frames). Each frame corresponds to three registered PNG images: RGB image, NIR image, and Level-3 semantic label map. The label map is a single-channel image with pixel values corresponding to category IDs (0-25). The directory structure is divided into train/ and val/, each containing three subdirectories: rgb/, nir/, and gt_labels/, organized by driveId and frame index. This dataset is suitable for safe semantic segmentation tasks in autonomous driving scenarios, especially for evaluating and improving the robustness of segmentation models under adverse weather conditions.
IDD-AW: 印度驾驶数据集——恶劣天气
数据集简介
IDD-AW(India Driving Dataset – Adverse Weather)是一个面向自动驾驶的语义分割基准数据集,专注于雨、雾、低光照和雪等恶劣天气条件下的场景理解。数据集提供配对的 RGB + 近红外(NIR) 图像以及密集的 Level-3 语义标签(共26个类别),旨在推动安全语义分割研究。
数据集规模
| 划分 | 驾驶序列数 | 帧数 |
|---|---|---|
| train | 143 | 3430 |
| val | 18 | 475 |
每帧包含三个对齐的 PNG 文件:RGB 图像、NIR 图像和 Level-3 标签图。数据集中已移除一帧无真值标签的训练帧(drive 175,frame 00000028),确保所有帧均具有完整配对。
目录结构
train/ , val/ rgb/<driveId>/<frame>_rgb.png # RGB 图像 nir/<driveId>/<frame>_nir.png # 近红外图像 gt_labels/<driveId>/<frame>_labellevel3Ids.png # Level-3 语义标签
<driveId>为整数文件夹名称(如58、175)<frame>为8位零填充索引(如00000028)- 标签图为单通道,像素值为类别ID
语义类别(Level-3,共26类)
| ID | 类别名称 | RGB 调色板 |
|---|---|---|
| 0 | road | (128, 64, 128) |
| 1 | drivable fallback | (81, 0, 81) |
| 2 | sidewalk | (244, 35, 232) |
| 3 | non drivable fallback | (152, 251, 152) |
| 4 | person | (220, 20, 60) |
| 5 | rider | (255, 0, 0) |
| 6 | motorcycle | (0, 0, 230) |
| 7 | bicycle | (119, 11, 32) |
| 8 | autorickshaw | (255, 204, 54) |
| 9 | car | (0, 0, 142) |
| 10 | truck | (0, 0, 70) |
| 11 | bus | (0, 60, 100) |
| 12 | vehicle fallback | (0, 0, 90) |
| 13 | curb | (196, 196, 196) |
| 14 | wall | (102, 102, 156) |
| 15 | fence | (190, 153, 153) |
| 16 | guard rail | (180, 165, 180) |
| 17 | billboard | (174, 64, 67) |
| 18 | traffic sign | (220, 220, 0) |
| 19 | traffic light | (250, 170, 30) |
| 20 | pole | (153, 153, 153) |
| 21 | obs-str-bar-fallback | (169, 187, 214) |
| 22 | building | (70, 70, 70) |
| 23 | bridge | (150, 100, 100) |
| 24 | vegetation | (107, 142, 35) |
| 25 | sky | (70, 130, 180) |
使用示例
下载数据集
python from huggingface_hub import snapshot_download
snapshot_download( "Furqan7007/IDDAW_OFFICIAL", repo_type="dataset", local_dir="IDDAW", )
加载单帧数据
python from pathlib import Path import numpy as np from PIL import Image
root = Path("IDDAW/train") rgb = np.array(Image.open(root / "rgb/58/00000000_rgb.png")) nir = np.array(Image.open(root / "nir/58/00000000_nir.png")) label = np.array(Image.open(root / "gt_labels/58/00000000_labellevel3Ids.png"))
引用信息
该数据集基于 WACV 2024 论文《IDD-AW: A Benchmark for Safe Semantic Segmentation in Adverse Weather》,引用格式: bibtex @inproceedings{iddaw2024, title = {IDD-AW: A Benchmark for Safe Semantic Segmentation in Adverse Weather}, author = {Shaik, Furqan Ahmed and others}, booktitle = {WACV}, year = {2024} }
其他信息
- 许可证:其他(需确认具体许可,参见 https://iddaw.github.io/)
- 任务类别:图像分割
- 标签:语义分割、自动驾驶、恶劣天气、近红外、安全分割
- 数据集规模:1K < N < 10K




