WatermarkRemoval

Watermark Removal

Deep Learning

install Watermark-Removal-Pytorch

$ git clone  https://github.com/braindotai/Watermark-Removal-Pytorch.git
$ cd Watermark-Removal-Pytorch
$ python3 -m venv .venv
$ source .venv/bin/activate
$ pip install -r requirements.txt

create mask

• at gimp (there might be better way)
  • crop watermark as tight (small) as possible
    • cropped hight needs at least 33 pixel (failed with 32)
    • cropped width should be smaller than 512(this is modifiable at interface.py)
  • make it gray scale
  • apply gamma to it max
  • make it b&w

do removal

$ python3 inference.py --training-steps 100 --image-path target-image.jpg --mask-path mask.jpg

TODO

• automatic mask creation
  • https://github.com/rohitrango/automatic-watermark-detection

References

• others
  • https://github.com/marcbelmont/cnn-watermark-removal

OpenCV

• Template Matching

  • sample application