CRAFT for text detection and location in images, implemented in python and pytorch
#ai #document-processing #text-detection #ocr #fer-r-and-d #Fern-R&D
boomb0om
20f0c94d6b
fix: fix model urls and polygons
|
пре 1 година | |
|---|---|---|
| CRAFT | пре 1 година | |
| images | пре 2 година | |
| .gitignore | пре 2 година | |
| README.md | пре 2 година | |
| example.ipynb | пре 2 година | |
| requirements.txt | пре 2 година | |
| setup.py | пре 2 година | |
| test.py | пре 2 година |
README.md
CRAFT-text-detection
An unofficial PyTorch implementation of CRAFT text detector with better interface and fp16 support
This is not official implementation. I partially use code from the original repository
Main features of this implementation:
- User-friendly interface
- Easier to integrate this model in your project
- fp16 inference support
- Automatically downloading weights from huggingface
Installation
Recommended:
pip install git+https://github.com/boomb0om/CRAFT-text-detection/
or
git clone https://github.com/boomb0om/CRAFT-text-detection
cd CRAFT-text-detection/
pip install -r requirements.txt
To test model you can run test.py file.
Examples
from PIL import Image
from CRAFT import CRAFTModel, draw_polygons
model = CRAFTModel('weights/', 'cuda', use_refiner=True, fp16=True)
img = Image.open('images/cafe_sign.jpg')
polygons = model.get_polygons(img)
result = draw_polygons(img, polygons)
You can find more usage examples in example.ipynb
Detected polygons:
