📌 ImageWatermark

This repository collects and encapsulates a variety of invisible image watermarking methods, including both generation-stage and post-processing watermarking. A unified interface is provided to easily embed and extract watermarks from images, enabling batch experiments, robustness evaluations, and comparative studies.

The currently supported watermarking methods are as follows:

Name	Type	Conference/Year	Paper
Hidden	Post-generation	ECCV 2018	HiDDeN: Hiding Data with Deep Networks
RivaGAN	Post-generation	ArXiv 2019	Robust Invisible Video Watermarking with Attention
Stable Signature	Post-generation	ICCV 2023	The Stable Signature Rooting Watermarks in Latent Diffusion Models
TreeRing	Noise selection	CVPR 2023	Tree-Ring Watermarks: Fingerprints for Diffusion Images that are Invisible and Robust
EditGuard	Post-generation Edit	CVPR 2024	EditGuard: Versatile Image Watermarking for Tamper Localization and Copyright Protection
PRC Watermark	Noise selection	ICLR 2025	An Undetectable Watermark for Generative Image Models
SleeperMark	Denoising process	CVPR 2025	SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models
dwtDctSvd	Post-generation	\	\
Stegastamp	Post-generation	CVPR 2020	Invisible Hyperlinks in Physical Photographs
Robust-wide	Post-generation	ECCV 2024	Invisible Hyperlinks in Physical PhotographsRobust-Wide: Robust Watermarking Against Instruction-Driven Image Editing
Smoothed HiDDeN	Post-generation	ECCV 2024	Certifiably Robust Image Watermark

🔧 Environment Setup

Two installation options are provided:

• Conda environment

  conda env create -f environment/environment.yml
  conda activate imagewatermarker
  

• Pip installation

  pip install -r environment/requirements.txt
  

---

📥 Pretrained Weights

Currently, pretrained weights are not hosted in this repository. Later I’ll provide a one-click download script. You can download them directly from the corresponding original repositories.

---

⚙️ Custom Configuration

All method-specific parameters, paths, and hyperparameters can be modified in the configs/ directory. Example:

# configs/smoothed_hidden.yaml
device: cuda
seed: 4
message_length: 30
options_file: ...
batch_size: 3
checkpoint_path: ...
message: "100100111111110010010000010111"
resize: 512
alpha: 0.001
sigma: 0.1
num_noise: 20
range: 0.4

🧪 Testing Methods

For example, to run Smoothed HiDDeN:

python -m watermarker.SmoothedHidden 

You can directly invoke any method or modify its implementation, using these modules as building blocks for larger systems