Practical overview
When it fits
Use it when building or studying a scene-image text-localization pipeline, particularly when long text predictions are a concern and a trainable Keras implementation is useful.
Who it is for
- Computer-vision developers adding scene-text localization to an image workflow
- Researchers or practitioners examining or training an EAST-derived text detector
Documented uses
- Locate text regions in scene images.
- Predict text-region boundaries as quadrilateral coordinates.
- Prepare labeled image data and train the detector.
- Run the prediction script on an image.
Requirements & limits
The code is released under the MIT License.
- The implementation uses Keras and lists Python 3.6.3+, TensorFlow 1.5.0+, Keras 2.1.4+, NumPy 1.14.1+ and tqdm 4.19.7.
- The repository documents both a training entry point and a prediction command, but does not detail pretrained-model installation.
- The author reports improved accuracy over EAST, especially on long text; the repository's comparison also notes limitations in EAST vertex-coordinate accuracy.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare data and start model training
Prepare the documented Tianchi ICPR training data and launch AdvancedEAST training.
Before you start
- Python 3.6.3 or later
- TensorFlow 1.5.0 or later, including tensorflow-gpu as an option
- Keras 2.1.4 or later, NumPy 1.14.1 or later, and tqdm 4.19.7
- Tianchi ICPR dataset with images and text files
Steps
Create a data root directory named icpr and place the dataset images and text files in it.
Adjust the training parameters in cfg.py.
Run preprocess.py to resize images, then run label.py to generate label information.
Run advanced_east.py as the training entry point.
Expected outcome
The repository's documented training process is started using the prepared dataset and configured model parameters.
Guide limitations
- The README points to dataset-format information and a dataset download; it does not detail dataset preparation beyond placing images and text files in the root directory.
- The guide covers the documented training stages and does not verify successful execution or model performance.
Run prediction on the documented demo image
Use the repository's documented prediction entry point with its demo image.
Before you start
- A configured Python environment with the listed TensorFlow, Keras, NumPy and tqdm dependencies
- The repository files and demo/001.png
- A pretrained model is linked in the README for testing
Steps
Obtain the repository and the pretrained model linked in its README.
Run the documented command: python predict.py -p demo/001.png.
Expected outcome
The documented prediction entry point is invoked for the demo image.
Guide limitations
- The README does not specify the pretrained model installation or placement procedure.
- This documents the command only; it does not establish successful execution or prediction quality.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- huoyijie/AdvancedEAST · Official homepage title · retained in its source language