Dataset details

The dataset consists of images collected in an unstructured road scenario, driving in adverse weather conditions of rain, fog, lowlight and snow. Each individual RGB image has a more detailed near-infrared image (NIR) captured simultaneously. The images are collected using JAI FS-3200D-10GE camera. The dataset comprises 5000 images, manually selected to represent various adverse weather scenarios, including rain, fog, low light, and snow. Each RGB image also has a paired NIR image to provide image enhancement. Each image is densely annotated at the pixel level for semantic segmentation, utilizing a label set with a hierarchical structure consisting of 7 labels at level 1 and 30 labels at level 4.

Test Dataset
IDD-AW Dataset page
Download here

Summary

Dataset Comparison

• IDD-AW provides more labeled images compared to other datasets.
• IDD-AW includes adverse weather conditions such as rain, fog, snow, and lowlight.
• IDD-AW offers a diverse range of labeled conditions, including NIR images, capturing unstructured driving environments.
• IDD-AW features 30 labels, providing comprehensive coverage of driving scenarios.
• Other datasets, such as ACDC, BDD100K, and WildDash, focus on specific adverse weather conditions or driving scenarios.

Our dataset annotations have unique labels like billboard, autorickshaw, animal etc. We also focus on identifying probable safe driving areas beside the road. The labels for the dataset are organized as a 4 level hierarchy. Unique integer identifiers are given for each of these levels. The histogram below gives:

Pixel counts for each label in the y axis.

The four level label hierarchy and the label ids for intermediate levels (level 2, level 3).

The color coding used for the prediction and ground truth masks are given to the corresponding bars.

Label Heirarchy and Statistics