├── Figure ├── Fig1.png ├── Fig2.png └── Fig3a.png └── README.md /Figure/Fig1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shanface33/AutoSplice_Dataset/HEAD/Figure/Fig1.png -------------------------------------------------------------------------------- /Figure/Fig2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shanface33/AutoSplice_Dataset/HEAD/Figure/Fig2.png -------------------------------------------------------------------------------- /Figure/Fig3a.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shanface33/AutoSplice_Dataset/HEAD/Figure/Fig3a.png -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # AutoSplice: A Text-prompt Manipulated Image Dataset 2 | 3 | [Shan Jia](https://shanface33.github.io/), 4 | Mingzhen Huang, Zhou Zhou, Yan Ju, Jialing Cai, [Siwei Lyu](https://cse.buffalo.edu/~siweilyu/)
5 | University at Buffalo, State University of New York, NY, USA
6 | 7 | The AutoSplice dataset was proposed in the [CVPR2023 Workshop on Media Forensics](https://sites.google.com/view/wmf2023/home) paper "AutoSplice: A Text-prompt Manipulated Image Dataset for Media Forensics", which leverages the DALL-E2 [^1] language-image model to automatically generate and splice masked regions guided by a text prompt. It consists of 5,894 manipulated and authentic images. 8 | 9 | [[Paper]](https://openaccess.thecvf.com/content/CVPR2023W/WMF/papers/Jia_AutoSplice_A_Text-Prompt_Manipulated_Image_Dataset_for_Media_Forensics_CVPRW_2023_paper.pdf) [[Download]](https://docs.google.com/forms/d/1bHbWZ-DsG1-VKaMs4Puy0996yj485x7HK13fgbNRerE/edit) 10 | 11 | ![fig1_compressed-1](Figure/Fig1.png) 12 | 13 | ## Summary 14 | The database contains 3, 621 images by locally or globally manipulating real-world images, and 2, 273 authentic images. 15 | 16 | Three JPEG compression versions along with their manipulation masks and captions are included, 17 | - JPEG-100, lossless compression with a JPEG quality factor of 100; 18 | - JPEG-90, gently lossy compression with a JPEG quality factor of 90; 19 | - JPEG-75, lossy compression with a JPEG factor of 75 (the same as the authentic images derived from the Visual News dataset[^2]), used for fine-tuning tasks. 20 | 21 | ![fig1_compressed-1](Figure/Fig2.png) 22 | [^1]: Ramesh, Aditya, et al. "Hierarchical text-conditional image generation with clip latents." arXiv preprint arXiv:2204.06125 (2022). 23 | [^2]: Liu, Fuxiao, et al. "Visual news: Benchmark and challenges in news image captioning." Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (2021). 24 | 25 | ## Comparison with Other Datasets 26 | ![fig1_compressed-1](Figure/Fig3a.png) 27 | 28 | ## Dataset Download 29 | The *AutoSplice* dataset can be downloaded through [Google Drive](https://drive.google.com/drive/folders/1QpBm4528ng877ytdBiRSnrH-rsrCeoIA?usp=sharing) or [Baidu Wangpan](https://pan.baidu.com/s/1SdbguPD2trFjmVT7iDnIyQ) (Password: rt97). If you have any questions, please send an email to [autosplice.dataset@gmail.com]. 30 | 31 | ## License and Citation 32 | The AutoSplice dataset is released only for academic research. Researchers from educational institutes are allowed to use this database freely for noncommercial purposes. 33 | 34 | If you use this dataset, please cite the following papers: 35 | ``` 36 | @inproceedings{jia2023autosplice, 37 | title={AutoSplice: A Text-prompt Manipulated Image Dataset for Media Forensics}, 38 | author={Jia, Shan and Huang, Mingzhen and Zhou, Zhou and Ju, Yan and Cai, Jialing and Lyu, Siwei}, 39 | booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, 40 | pages={893--903}, 41 | year={2023} 42 | } 43 | --------------------------------------------------------------------------------