Image Text Recognition

Recognite text from image with CRNN network.

Install requirements

pip install -r requirements.txt

Dataset

Full dataset used is Synthetic Word Dataset
Dataset used for project here.

Model's weights

Model's weights for prediction and evaluation model.weights.hdf5

Download data and model's weights

cd text-recognition/
python lib/preprocess/download_data.py

Extract data and model's weights

tar -xzvf data.tar.gz
rm -rf data.tar.gz

Create checkpoints folder

mkdir Data/checkpoints/

Run application

python main.py

Application run on http://localhost:8096

Api

1. Training model

   • URL: /train
   • Method: GET
   • Usage: Training model and save weights

2. Evaluation

   • URL: /evaluation
   • Method: POST
   • Body:
     • filename: Path to file contain image's paths and labels
     • paths: List of image's paths
     • labels: List of image's labels
   • Usage: Evaluate accuracy and letter accuracy of model with data described in filename or (paths, labels).
   • Return: Accuracy, letter accuracy

3. Prediction

   • URL: /prediction
   • Method: POST
   • Body:
     • img: Path to image for prediction
   • Usage: Predict text in new image
   • Return: Text predicted