Skip to content

Latest commit

 

History

History
60 lines (49 loc) · 4.04 KB

README.md

File metadata and controls

60 lines (49 loc) · 4.04 KB

Rethinking Image Editing Detection in the Era of Generative AI Revolution

The Generative Regional Editing (GRE) dataset was proposed in the MM2024 paper "Rethinking Image Editing Detection in the Era of Generative AI Revolution".

Dataset

Introduction

Considering copyright issues, the GRE dataset only provides edited images. The original images were collected from datasets such as COCO, Flickr2k, and VisualNews, as detailed in Section 3.1 Original Image Collection of the paper.

Information and labels related to generative editing methods are included in the file name xxxxxxxxxxxx.jpg. The parsing code is as follows:

file_name = '010502003a25.jpg'
edit_method_tag = int(file_name[2:4])

edit_method_dict = {
    1: 'GAN-MAT',
    2: 'GAN-LaMa',
    3: 'SD-Stable Diffusion V2.0',
    4: 'SD-ControlNet',
    5: 'SD-PaintByExample',
    6: 'Software-PhotoShop',
    7: 'Online-Weibo',
    8: 'Online-X(Twitter)',
}
edit_method = edit_method_dict[edit_method_tag]

Download

If you would like to access the GRE dataset, please fill out this Google Form. The download link will be sent to you once the form is accepted.

Authentic Subset

We select five authentic image datasets in the two most frequently tampered or edited scenarios: Daily Moment Snapshots (COCO, Flickr2k, DIV2k, SR-RAW) and News & Public Sentiment Visuals (VisualNews). These images also constitute the authentic subset. Due to copyright reasons, we did not incorporate these datasets into GRE. However, for your convenience, we provide the official (or unofficial) dataset download links in the table below.

Dataset Paper Download URL
COCO2017 Microsoft coco: Common objects in context https://cocodataset.org/#download
Flickr2K (HR images) - https://github.com/LimBee/NTIRE2017
[UNOFFICIAL]https://www.kaggle.com/datasets/daehoyang/flickr2k
DIV2K (HR images) - https://data.vision.ee.ethz.ch/cvl/DIV2K/
SR-RAW Zoom to learn, learn to zoom https://drive.google.com/drive/folders/1FHhcrZjYvFm-zliziQIRVzYjlZUCikai
VisualNews (images) Visual news: Benchmark and challenges in news image captioning https://www.cs.rice.edu/~vo9/visualnews/

In addition, we also provide a small number of images from real scenes collected online, which can be downloaded separately here and is also included in the download link you requested through the Google Form. Once again, these images are collected from the Internet and are only used for non-commercial purposes.

License and Citation

The GRE dataset is released only for academic research. Researchers from educational institutes are allowed to use this database freely for non-commercial purposes.

Reference Format:

@inproceedings{10.1145/3664647.3681445,
    author = {Sun, Zhihao and Fang, Haipeng and Cao, Juan and Zhao, Xinying and Wang, Danding},
    title = {Rethinking Image Editing Detection in the Era of Generative AI Revolution},
    year = {2024},
    publisher = {Association for Computing Machinery},
    address = {Melbourne, VIC, Australia},
    url = {https://doi.org/10.1145/3664647.3681445},
    doi = {10.1145/3664647.3681445},
    booktitle = {Proceedings of the 32nd ACM International Conference on Multimedia},
    series = {MM '24}
}