EFFECTIVE IMAGE TAMPERING LOCALIZATION VIA ENHANCED TRANSFORMER AND CO-ATTENTION FUSION

An official implementation code for paper "Effective Image Tampering Localization via Enhanced Transformer and Co-attention Fusion". This repo provide codes, pretrained/trained weights, and our training datasets.

Network Architecture

Update

• 24.05.18. We upload the training dataset.
• 24.04.26. We updated the weight which can be downloaded from Google Drive Link or Baiduyun Link (password：EITL) and the file nets/EITLnet.py. The latest corrected experimental results are marked in red in the table below, which the average performance is more higher than before(paper).

Environment

• Python 3.8
• cuda11.1+cudnn8.0.4

Requirements

• pip install -r requirements.txt

Training dataset

The training dataset catalog is as follows.

├─train_dataset
    ├─ImageSets
    │  └─Segmentation
    │          train.txt
    │          val.txt
    ├─JPEGImages
    │      00001.jpg
    │      00002.jpg
    │      00003.jpg     
    │      ...
    └─SegmentationClass
            00001_gt.png
            00002_gt.png
            00003_gt.png

Please download the training dataset from Kaggle Link or Baiduyun Link(password:EITL). The SegmentationClass0255 folder contains ground truth values of only 0 and 255, while the SegmentationClass01 folder contains values of only 0 and 1. In this paper, the latter folder was used for training. The SegmentationClass01 folder was generated from the former using the to01.py script.

Trained Models

Please download the weight from Google Drive Link or Baiduyun Link(password:EITL) and place it in the weights/ directory.

Training

python train.py

Testing

python test.py

Bibtex

@inproceedings{guo2023effective,
 title={Effective Image Tampering Localization via Enhanced Transformer and Co-attention Fusion},
 author={Guo, Kun and Zhu, Haochen and Cao, Gang},
 booktitle={ICASSP},
 year={2024}
}

Contact

If you have any questions, please ask in the issue.