Towards Generic Image Manipulation Detection with Weakly-Supervised Self-Consistency Learning
June 28, 2024 · View on GitHub
Towards Generic Image Manipulation Detection with Weakly-Supervised Self-Consistency Learning
Yuanhao Zhai, Tianyu Luan, David Doermann, Junsong Yuan
University at Buffalo
ICCV 2023
Our WSCL (weakly-supervised self-consistency learning) can detect and localize image manipulations, using only image-level binary labels for training.
This repo contains the MIL-FCN version of our WSCL implementation.
🚨News
03/2024: add demo script! Check here for more details, and check here for the online Gradio demo!
1. Setup
Clone this repo
git clone git@github.com:yhZhai/WSCL.git
Install packages
pip install -r requirements.txt
2. Data preparation
We provide preprocessed CASIA (v1 and v2), Columbia, and Coverage datasets here.
Place them under the data folder.
For other datasets, please prepare a json datalist file with similar structure as the existing datalist files in the data folder. After that, adjust the train_dataslist or the val_datalist entries in the configuration files configs/final.yaml.
3. Training and evaluation
Runing the following script to train on CASIAv2, and evalute on CASIAv1, Columbia and Coverage.
python main.py --load configs/final.yaml
For evaluating a pre-trained checkpoint:
python main.py --load configs/final.yaml --eval --resume checkpoint-path
We provide our pre-trained checkpoint here.
4. Demo
Running our manipulation model on your custom data!
Before running, please configure your desired input and output path in the demo.py file.
python demo.py --load configs/final.yaml --resume checkpoint-path
By default, it evaluates all .jpg files in the demo folder, and saves the
detection result in tmp, with manipulation probablities appended to the file names.
Citation
If you feel this project is helpful, please consider citing our paper
@inproceedings{zhai2023towards,
title={Towards Generic Image Manipulation Detection with Weakly-Supervised Self-Consistency Learning},
author={Zhai, Yuanhao and Luan, Tianyu and Doermann, David and Yuan, Junsong},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={22390--22400},
year={2023}
}
Acknowledgement
We would like to thank the following repos for their great work: