text_detection_tech

text detection MMOCR

Install MMOCR

conda create -n open-mmlab python=3.8 pytorch=1.10 cudatoolkit=11.3 torchvision -c pytorch -y
conda activate open-mmlab
pip3 install openmim
git clone https://github.com/open-mmlab/mmocr.git
cd mmocr
mim install -e .

Inference

It is recommended to use TextSnake pretrained model for good defaults result

text detection PaddleOCR

PaddleOCR installation

clone PaddleOCR https://github.com/PaddlePaddle/PaddleOCR

Install paddle lib on conda https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/conda/linux-conda.html

conda install paddlepaddle-gpu==2.2.2 cudatoolkit=11.2 -c https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/Paddle/ -c conda-forge

Image label format

preprocess folder contain script that convert ICDAR and CTPN format to paddleOCR format
ICDAR format

"polygon_points,lang,text"
78,55,419,55,419,109,78,109,Latin,###
111,283,1521,283,1521,323,111,323,Latin,###

text-detection-CTPN format

"(xmin, ymin, xmax, ymax)"
28,20,31,40
32,20,47,40
48,20,63,40
64,20,79,40
80,20,95,40

PaddleOCR format

" Image file name             Image annotation information encoded by json.dumps"
img_file_name.jpg    [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]

if you plan on training text detection only, change transcription to random text like AAAA

Training data in yml

data_dir: path_to_folder_img
label_file_list: path_to_label_file.txt

final image path wil be joined with data_dir variable in yml file

data_dir + img_file_name.jpg

Training

Detail for customizing PaddleOCR

config PaddleOCR config
configs folder contain 1 sample config file

Training scripts with config

modify the yml file with training data to quickly train the model

python3 tools/train.py -c configs/det/det_mv3_db_dk.yml

-o to modify the yml variable without edit it

It is recommended to run the defaults model on new data and modify those data, then train with the modified data 500 data samples, lr=0.001, epoch=500 is recommended for quick training with acceptable result

Convert trained model to inference model

python3 tools/export_model.py -c configs/det/det_mv3_db_dk.yml -o Global.pretrained_model="./output/det_db/best_accuracy"

change Global.pretrained_model variable to the just trained models path

Prediction

Use the inference model to get prediction result

command line

python3 tools/infer/predict_det.py --det_algorithm="DB" --det_model_dir="./output/det_db_inference/" --image_dir="./doc/imgs/" --use_gpu=True

code

custom_predict.py

text_detection Class

CustomPaddleOCR.py inherit from main PaddleClass
text_detector_paddle.py add custom pre/post_process text detection