hongleizhang / IGA

The source code for the paper "Robust Data Hiding Using Inverse Gradient Attention".

IGA

The source code for the paper "Robust Data Hiding Using Inverse Gradient Attention".

https://arxiv.org/abs/2011.10850

Installation

Python 3.6.7

torch 1.0.0

torchvision 0.2.1

numpy 1.19.4

Data Preparation

For coco dataset, we use 10,000 images for training and 1,000 images for validation. Thus, we chose those 10,000 + 1,000 images randomly from one of the coco datasets. http://cocodataset.org/#download.

For DIV2K dataset, we use 800 images for training and 100 images for validation, and it can be downloaded from https://data.vision.ee.ethz.ch/cvl/DIV2K/.

The data directory has the following structure:

<data_root>/
  train/
    train_class/
      train_image1.jpg
      train_image2.jpg
      ...
  val/
    val_class/
      val_image1.jpg
      val_image2.jpg
      ...

train_class and val_class folders are so that we can use the standard torchvision data loaders without change.

Model Running

By default, you can run iga with identity settings. For instance,

python -u main.py new -d ../data_path/coco -b 32 -m 90 -r 30 --name iga_identity

You can also run iga with combined noises settings. For instance,

python -u main.py new -d ../data_path/coco -b 32 -m 90 -r 30 --noise "crop((0.4,0.55),(0.4,0.55))+cropout((0.25,0.35),(0.25,0.35))+dropout(0.25,0.35)+resize(0.4,0.6)+jpeg()" --name iga_cn

Citation

If this repository is useful for your research, please consider citing our paper:

@article{zhang2020iga,
  title     = {Robust Data Hiding Using Inverse Gradient Attention},
  author    = {Honglei Zhang, Hu Wang, Yuanzhouhan Cao, Chunhua Shen and Yidong Li},
  journal   = {arXiv preprint arXiv:2011.10850},
  year      = {2020}
}