CUDA Blind Watermark

Quickstart

1. Clone this repository and enter.
2. Build the project (make sure OpenCV and CUDAToolkit can befound by cmake)

mkdir build && cd build
cmake ..
make

3. Play with the CLI

./WmCLI
> embed pic_path watermark_path output_path
> extract wm_rows wm_cols pic_path output_path

Test

All test scripts can be found in ./demo

• test_gpu.py: A wrapper for WmCLI to test all the images in ./demo/pic
• test_cpu.py: A wrapper for guofei-wm to test all the images in ./demo/pic
• test_video.py: Additional expiriement for video watermark embedding
• test_tool.py: Tools for attacking the image

The scripts can be simply run as:

mkdir ./demo/out \
      ./demo/out/attack \
      ./demo/out/embeded \
      ./demo/out/extracted \
      ./demo/out/thresh
python test_gpu.py