AODRaw: Towards RAW Object Detection in Diverse Conditions

Dataset Overview

AODRaw dataset contains 7,785 high‑resolution real RAW images, with 135,601 annotated instances covering 62 categories, capturing indoor and outdoor scenes under nine different lighting and weather conditions. The dataset supports RAW and sRGB object detection and provides a comprehensive benchmark to assess current detection methods.

Dataset Structure

The dataset is organised into several directories, each containing images processed in different ways and annotation files.

Directory Structure

├── AODRaw ├── images (435G) ├── images_downsampled_raw (223G) ├── images_downsampled_srgb (4.3G) ├── images_slice_raw (439G) ├── images_slice_srgb (23G) ├── annotations

Directory Purpose

• images: Original RAW and sRGB images, resolution $6000\times4000$, total 7,785 images.
• images_downsampled_raw: Down‑sampled RAW images, resolution $2000\times1333$, total 7,785 images.
• images_downsampled_srgb: Down‑sampled sRGB images, resolution $2000\times1333$, total 7,785 images.
• images_slice_raw: Sliced RAW images, resolution $1280\times1280$, total 71,782 images.
• images_slice_srgb: Sliced sRGB images, resolution $1280\times1280$, total 71,782 images.
• annotations: Annotation files, containing training and testing annotation information.

Download Links

• images: Baidu, TeraBox
• images_downsampled_raw: Baidu
• images_downsampled_srgb: Baidu, TeraBox
• images_slice_srgb: Baidu, TeraBox
• annotations: Baidu, Google, TeraBox

Dataset Information

Images

Annotations

Annotations follow the COCO format, containing image IDs, file names, height, width, and information on lighting and weather conditions.

Citation

@article{li2024aodraw, title={Towards RAW Object Detection in Diverse Conditions}, author={Zhong‑Yu Li and Xin Jin and Boyuan Sun and Chun‑Le Guo and Ming‑Ming Cheng}, journal={arXiv preprint arXiv:2411.15678}, year={2024}, }

License

Code is released under the Creative Commons Attribution‑NonCommercial‑ShareAlike 4.0 International Public License for non‑commercial use only.