Auto Text Removal

Auto text removal using CRAFT to detect the text and LaMa to remove the text. Highly motivated by auto-lama which do object detection + LaMa.

Implementation pipeline

Applying LaMa multiple times with bigger mask creates a better result.

How to use it

1. Clone this repository
2. Setup and download CRAFT model. Please see the official repo for the detail. Please put the downloaded model under CRAFT-pytorch folder.
3. Setup and download LaMa model. Please see the official repo for the detail. Please put the downloaded model under LaMa folder.
4. Put your input images to images folder
5. Run the program

bash script/predict_and_inpaint.sh

or

bash script/predict_and_inpaint.sh 5

indicating repeat LaMa 5 times (default 3 times)

Output will be located at results folder and result from every repetition will located at each_step_results folder

Result

Before (Photo source) After

Note

I do a little modification to CRAFT-pytorch original code to fit my usage requirement.

Contributor

Anyone who want to help is welcomed :D. Just open an issue or a PR.

Acknowledgements

• CRAFT-pytorch for text detection
• LaMa for object(text) removal

License

I am not really clear about the license I should use for this repo. https://github.com/advimman/lama use Apache-2.0 license while https://github.com/clovaai/CRAFT-pytorch use MIT license. Feel free to open an issue if you think the license that I use (MIT license) to label this repo is incorrect.