OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

East

EAST is a TensorFlow implementation of a scene-text detector that identifies text regions in images and returns their locations as boxes. The repository provides pretrained RBOX weights, evaluation and training code, a local web demo, and a C++ locality-aware non-maximum suppression component.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

East — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use EAST when a project needs scene-text localization from images and can work with its TensorFlow implementation and RBOX output.

Who it is for

  • Developers adding text-region localization before OCR.
  • Researchers or engineers evaluating scene-text detection on image collections.
  • Teams preparing annotated image datasets to train a scene-text detector.

Documented uses

  • Detect and localize text in images for an OCR pipeline.
  • Evaluate image sets and write detection results to an output directory.
  • Train the detector using image-specific ground-truth text files.
  • Run a local web demo to submit images and inspect detected boxes.
Before you choose

Requirements & limits

The repository is public and specifies the GPL-3.0 license.

  • The implementation supports RBOX geometry; the README says only the RBOX part is implemented.
  • Training data needs one ground-truth text file per image; ICDAR 2015 labels may require filename changes and removal of extra characters.
  • The README reports performance figures for particular hardware and NMS implementations, so they should be read in that context.
  • The hosted demo is described as running on one CPU core, and simultaneous use can degrade response time.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up EAST’s local image demo

Start the documented web demo using a downloaded pretrained checkpoint.

Before you start

  • TensorFlow version above 1.0.
  • A downloaded EAST pretrained model checkpoint.

Steps

  1. Download a pretrained EAST model.

  2. Run the command below with the checkpoint path adjusted to its location.

    Documented command
    python3 run_demo_server.py --checkpoint-path /path/to/checkpoint

  3. Open http://localhost:8769 to use the web demo.

Expected outcome

A local web interface for submitting images and viewing detected text boxes.

Guide limitations

  • The demo result URL changes after an image is submitted.
  • Shared result URLs persist only while the corresponding data remains in static/results.
Documentation 1 ↗
Source-based practical guide

Prepare the documented custom-training inputs

Prepare the dataset and pretrained checkpoint needed to start EAST training.

Before you start

  • A dataset directory with a separate ground-truth text file for each image.
  • A pretrained ResNet-50 checkpoint.
  • TensorFlow version above 1.0.

Steps

  1. Place the training images and their ground-truth text files in the dataset path.

  2. For ICDAR 2015 data, rename ground-truth files from gt_img_*.txt to img_*.txt, or change the relevant code in icdar.py.

  3. Remove the extra characters noted in the README and review the examples in training_samples/.

Expected outcome

Dataset labels and model checkpoint are prepared for the repository’s documented training command.

Guide limitations

  • This guide covers input preparation only; it does not verify training execution or model results.
  • The repository’s training instructions require dataset-specific paths and settings.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
argman/EAST · Official homepage title · retained in its source language
How we review sourcesSuggest a correction