B09902005/scene_text_removal

Structure of this project:

Folders that put images:

• input : input image
• mask1 : mask of the 1st epoch
• input2 : input of the 2nd epoch (also output of ist epoch)
• mask2 : mask of the 2nd epoch
• output : final output

Folders that put intermediates:

• CRAFT_result : results of CRAFT
• segmentation_of_input : results of segment-anything with input
• segmentation_of_input2 : results of segment-anything with input2

Folders that contain codes:

• CRAFT-pytorch : The model to find words. (https://github.com/clovaai/CRAFT-pytorch)
• lama : The model to clean words. (https://github.com/advimman/lama)
• segment-anything : The model to do segmentation. (https://github.com/facebookresearch/segment-anything)

Bash files to process:

• run1.sh : For 1st epoch of lama
• run2.sh : For segmentation and mask generation for 2nd epoch
• run3.sh : For 2nd epoch of lama

Flowchart:

• flowchart.png : shows the structure of this project

How to use this:

• download necessary packages according to the github links of the folders above (for lama environment, use virtualenv instead of conda (there will be a folder called lama/inpenv) )
• download CRAFT model from https://drive.google.com/file/d/1Jk4eGD7crsqCCg9C9VjCLkMN3ze8kutZ/view and put it in CRAFT-pytorch/weights
• download lama model from https://huggingface.co/smartywu/big-lama/resolve/main/big-lama.zip and unzip it as lama/big-lama
• download segment-anything model from https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth and put it in segment-anything/
• Run this code

bash run1.sh
bash run2.sh
bash run3.sh