cs-chan/Total-Text-Dataset

Total Text Dataset. It consists of 1555 images with more than 3 different text orientations: Horizontal, Multi-Oriented, and Curved, one of a kind.

What it solves

This project provides a specialized dataset for the detection of curved text in natural scenes, addressing the challenge of identifying text that does not follow a straight horizontal line (arbitrary-shaped text).

How it works

Total-Text is a word-level English curved text dataset. It provides groundtruth data in multiple formats, including text-level and pixel-level annotations, to help models learn to detect non-linear text. The repository also includes a baseline implementation (Poly-FRCNN-3), a Guided Annotation toolbox for creating scene text image annotations, and an evaluation script (DetEval.py) to benchmark detection performance using precision, recall, and F-score.

Who it’s for

It is designed for computer vision researchers and developers working on scene text detection, specifically those focusing on arbitrary-shaped or curved text reading tasks.

Highlights

  • Curved Text Focus: Specifically targets word-level English text that is curved or arbitrarily shaped.
  • Comprehensive Groundtruth: Offers both text-level and pixel-level groundtruth data.
  • Annotation Tools: Includes a dedicated Guided Annotation toolbox for scene text image annotation.
  • Benchmarking: Maintains a detailed detection leaderboard comparing various state-of-the-art methods.

Related

  • Project
    xiaofengShi/CHINESE-OCR

    A comprehensive pipeline for Chinese OCR in natural scenes, combining text orientation detection, CTPN for region detection, and CRNN for end-to-end character recognition.

  • Project
    wkentaro/labelme

    labelme is a cross‑platform Python/Qt desktop app for drawing bounding boxes, polygons, masks, etc., on images or video. Annotations are saved as JSON and can be exported to VOC or COCO formats, making it a practical tool for creating training data for computer‑vision models. Install via pip, a paid standalone executable, or Linux packages; recent versions add AI‑assisted mask generation (SAM, YOLO‑World).

  • Project
    inception-project/inception

    A machine-assisted text annotation platform for building annotated corpora with multi-layer schemes and knowledge-base entity linking.

  • Project
    opendatalab/OmniDocBench

    A comprehensive benchmark for evaluating document parsing in real-world scenarios, featuring rich annotations for layout detection, OCR, table, and formula recognition.

  • Project
    jolibrain/deepdetect

    A deep learning runtime and REST server that provides a unified API for training and inference across images, text, and tabular data.