机器学习与语料Free to access
Key facts
Contents & fields
The dataset contains 10,000 RGB images and corresponding annotation files organized in the following folders:
- RGB images——RGB images extracted from 30fps video at one frame per second, 10,000 images indexed sequentially.
- 2d box annotations——.txt files per image with object label, x-y coordinates, width, and height.
- Motion instance annotations——Pixel-wise masks for moving objects to be masked.
- Calibration——Single JSON file with intrinsic and extrinsic camera parameters.
- Ground Truth——Subfolder under Motion instance annotations with correct labels as images.
Research uses
Suitable for computer vision tasks in autonomous driving with fisheye cameras, such as object detection, moving object segmentation, depth estimation, and 3D reconstruction, especially for model adaptation in non-uniform road environments.
This card was drafted from the source page; institution, coverage, time span, scale, fields and license are subject to the official page (pending human review).
Keywords
Why this is hard to get on your own
Fisheye camera autonomous driving datasets are scarce, and existing ones mostly from developed countries' regulated roads, lacking annotated data for Indian semi-urban undivided roads.