https://github.com/18520339/dbnet-tf2
A TensorFlow 2 reimplementation of DBNet available as a Python package for Scene Text Detection, following ICDAR 2015 Dataset format and using TedEval as Evaluation metrics
https://github.com/18520339/dbnet-tf2
dbnet differentiable-binarization icdar ocr scene-text-detection tensorflow
Last synced: about 2 months ago
JSON representation
A TensorFlow 2 reimplementation of DBNet available as a Python package for Scene Text Detection, following ICDAR 2015 Dataset format and using TedEval as Evaluation metrics
- Host: GitHub
- URL: https://github.com/18520339/dbnet-tf2
- Owner: 18520339
- License: mit
- Created: 2022-06-02T23:59:58.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2023-05-10T20:59:16.000Z (about 2 years ago)
- Last Synced: 2025-03-16T20:55:32.710Z (2 months ago)
- Topics: dbnet, differentiable-binarization, icdar, ocr, scene-text-detection, tensorflow
- Language: Python
- Homepage: https://pypi.org/project/tfdbnet
- Size: 2.1 MB
- Stars: 5
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# TFDBNet
A TensorFlow 2 reimplementation of [Real-time Scene Text Detection with Differentiable Binarization](https://arxiv.org/abs/1911.08947) available as a Python package and using [TedEval](https://github.com/clovaai/TedEval) for evaluation metrics.
## Data Preparation
Store images in `imgs` folder and groundtruths in `gts` folder. Then, prepare text files for training and validate data in the following format with '\t' as a separator:
- Example for ICDAR 2015 `train.txt`:
```
./datasets/train/train_imgs/img_1.jpg ./datasets/train/train_gts/gt_img_1.txt
./datasets/train/train_imgs/img_2.jpg ./datasets/train/train_gts/gt_img_2.txt
```
- Example for ICDAR 2015 `validate.txt`:
```
./datasets/validate/validate_imgs/img_1.jpg ./datasets/validate/validate_gts/gt_img_1.txt
./datasets/validate/validate_imgs/img_2.jpg ./datasets/validate/validate_gts/gt_img_2.txt
```
You can customize the script in [dir2paths.sh](dir2paths.sh) to generate the above `train.txt` and `validate.txt` for your own dataset. And the groundtruths can be `.txt` files, with the following format:
```
x1,y1,x2,y2,x3,y3,x4,y4,annotation
```
Below is the content of `./datasets/train/train_gts/gt_img_1.txt`:
```
377,117,463,117,465,130,378,130,Genaxis Theatre
493,115,519,115,519,131,493,131,[06]
374,155,409,155,409,170,374,170,###
492,151,551,151,551,170,492,170,62-03
376,198,422,198,422,212,376,212,Carpark
494,190,539,189,539,205,494,206,###
374,1,494,0,492,85,372,86,###
```
## Quick Start
```
pip install tfdbnet
```
After installation, see the [demo](demo/demo.ipynb) on ICDAR 2015 dataset to know how to use. You can download my example trained weights along with the 2 files `train.txt` and `validate.txt` mentioned above [here](https://drive.google.com/file/d/1rLZiOTwlWtnq_a0v_oa0_2tSg7Mt1CUF).
## Reference
- https://github.com/MhLiao/DB
- https://github.com/zonasw/DBNet
- https://github.com/WenmuZhou/DBNet.pytorch
- https://github.com/xuannianz/DifferentiableBinarization
- https://github.com/clovaai/TedEval