10,000 images from 4 surround-view fisheye cameras providing 360-degree coverage for autonomous driving tasks. The dataset includes annotations for semantic segmentation, 2D/3D bounding boxes, and lens soiling detection.
Use Cases
- Train semantic segmentation models using the 40 class pixel-level masks to identify road, lanemark, and curb features
- Develop object detection algorithms for fisheye imagery using the 2D and 3D bounding box annotations for vehicles and pedestrians
- Build lens soiling detection systems using the binary soiling labels to trigger camera cleaning mechanisms
Strengths
- 10,000 images captured from four synchronized fisheye cameras (front, rear, left, right)
- Pixel-level semantic segmentation masks for 40 different classes including road, vehicle, and pedestrian
- Includes lens soiling annotations to detect occlusion from dirt or water on the camera lens
- Provides 2D and 3D bounding box labels for object detection in distorted fisheye coordinates