Fighting Fake News: Image Splice Detection via Learned Self-Consistency
[paper] [website]
Minyoung Huh *12, Andrew Liu *1, Andrew Owens1, Alexei A. Efros1
In ECCV 2018.
UC Berkeley, Berkeley AI Research1
Carnegie Mellon University2
Abstract
In this paper, we introduce a self-supervised method for learning to detect visual manipulations using only unlabeled data. Given a large collection of real photographs with automatically recorded EXIF meta-data, we train a model to determine whether an image is self-consistent -- that is, whether its content could have been produced by a single imaging pipeline.
1) Prerequisites
First clone this repo
git clone --single-branch https://github.com/minyoungg/selfconsistency
All prerequisites should be listed in requirements.txt. The code is written on TensorFlow and is run on Python2.7, we have not verified whether Python3 works. The following command should automatically load any necessary requirements:
bash pip install -r requirements.txt
2) Downloading pretrained model
To download our pretrained-model run the following script in the terminal:
chmod 755 download_model.sh && ./download_model.sh
3) Demo
To run our model on an image run the following code:
python demo.py --im_path=./images/demo.png
We also provide a normalized cut implementation by running the code:
python ncuts_demo.py --im_path=./images/ncuts_demo.png
We have setup a ipython notebook demo here
Disclaimer: Our model works the best on high-resolution natural images. Frames from videos do not generally work well.
Citation
If you find our work useful, please cite:
@inproceedings{huh18forensics,
title = {Fighting Fake News: Image Splice Detection via Learned Self-Consistency}
author = {Huh, Minyoung and Liu, Andrew and
Owens, Andrew and Efros, Alexei A.},
booktitle = {ECCV},
year = {2018}
}
Questions
For any further questions please contact Minyoung Huh or Andrew Liu
3 Oct 31, 2022
4 Jul 27, 2022
1 Nov 12, 2021
46 Dec 24, 2022
11 Dec 15, 2022
56 Dec 22, 2022
132 Dec 21, 2022
304 Jan 3, 2023
407 Dec 30, 2022
90 Oct 19, 2022
163 Dec 4, 2022
96 Nov 30, 2022
39 Dec 24, 2022
42 Dec 9, 2022
25 Sep 6, 2022
118 Jan 6, 2023
71 Nov 29, 2022
32 Nov 17, 2022