watermark-removal - Machine Learning-Based Image Processing for Invisible Watermark Elimination

Watermark-Removal Project

The Watermark-Removal project is an open-source initiative designed to eliminate watermarks from images using advanced machine learning techniques. This innovative project ensures that the modified images are indistinguishable from their original, watermark-free versions.

Inspiration and Foundation

The project draws inspiration from two significant research papers:

Contextual Attention - CVPR 2018
Gated Convolution - ICCV 2019

These papers explore deep image inpainting methods, which have laid the groundwork for developing this watermark removal tool. Chimzuruoke Okafor, the project's creator, expresses gratitude to Chu-Tak Li for his insightful Medium article series, which provided significant understanding into these key academic works.

Key Features

Machine Learning-Based: Utilizes image inpainting technology to seamlessly remove watermarks.
Indistinguishability: Ensures the processed image is visually identical to the original watermark-free image.
Open Source: Freely available for anyone interested in improving or adapting the project.

How to Run the Project

To start using the Watermark-Removal project, follow these steps:

Google Colab: This platform is recommended for running the code due to its simplicity and accessibility.

Clone the Repository:

!git clone https://github.com/zuruoke/watermark-removal

Navigate to the Directory:
```
!cd watermark-removal
```
Downgrade Tensorflow: Google Colab might use a newer Tensorflow version, but this project requires version 1.15.0:
```
!pip install tensorflow==1.15.0
```

Install Neuralgym:

!pip install git+https://github.com/JiahuiYu/neuralgym

Model Setup: Download the necessary model directories using the provided link and place them in the model/ folder. Be sure to rename checkpoint.txt to checkpoint.

Remove Watermark: Execute the main script to remove a watermark from an image:

!python main.py --image path-to-input-image --output path-to-output-image --checkpoint_dir model/ --watermark_type istock

The Watermark-Removal project is a powerful tool for anyone looking to restore images to their original state by removing obstructive watermarks effortlessly using state-of-the-art machine learning technology.