Adaptive Robust Watermarking Method Based on Deep Neural Networks

Dependencies and Installation

• Python 3 (Recommend to use Anaconda).
• tensorflow = 1.1.14 .
• Download dependencies

# 1. 克隆环境
conda env create -f TF114.yaml

# 2. 安装依赖
pip install -r requirements.txt

Get Started

• Run python main.py for training.

# Recommended training methods for embedding 100 bit messages
python main.py --exp_name wm100 --secret_len 100 --cover_h 400 --cover_w 400 --num_epochs 200 --batch_size 4 --lr .0001 --dataset_path /home/Dataset/train/mirflickr --loss_lpips_ratio 1.5 --loss_mse_ratio 2 --loss_secret_ratio 3.5 --GPU 0 --damping_end 0.2

Dataset

• In this paper, we use the commonly used dataset Mirflickr, and ImageNet.

• For train on your own dataset, change the code in main.py:

  line23: --dataset_path = 'your dataset'

Tensorboard

# Monitoring with tensorboard
cd code path
tensorboard --logdir ./logs --port 6006

result

ACC:

PSNR:

LOSS: