This repo contains the official codes for our paper:

Accepted in IEEE Transactions on Information Forensics and Security.

To test the model, simply run infer.py. It will probe the images in the results folder.

python infer.py

python evaluate.py

For training, you need to first download the training dataset and put them into the dataset folder correspondingly.

All file paths about training are stored in config.py. You may modify them to accommodate your system. Besides, the weights of ACBG are stored in the ckpt folder.

• Baidu Netdisk (百度网盘)：https://pan.baidu.com/s/1UgzcIo-Vpg2GsrfyjV-mjA?pwd=7yxr
• Extraction Code (提取码)：7yxr

If ACBG helps your research or work, please kindly cite our paper. The following is a BibTeX reference.

@ARTICLE{10589438,
  author={Liu, Wenxi and Zhang, Hao and Lin, Xinyang and Zhang, Qing and Li, Qi and Liu, Xiaoxiang and Cao, Ying},
  journal={IEEE Transactions on Information Forensics and Security}, 
  title={Attentive and Contrastive Image Manipulation Localization With Boundary Guidance}, 
  year={2024},
  volume={19},
  number={},
  pages={6764-6778},
  keywords={Location awareness;Contrastive learning;Feature extraction;Task analysis;Decoding;Visualization;Deepfakes;Image manipulation detection/localization},
  doi={10.1109/TIFS.2024.3424987}}