Introducing

English | chinese version

ToyOCR is an open-source toolbox based on PyTorch and detectron2 for text detection, text recognition and layout_analysis.

Requirements

• Python >= 3.7
• PyTorch >= 1.4
• torchvision
• OpenCV
• pycocotools

Main features

-Simple but not easy the structure is simple,the training speed is fast and the deploying is convenient.

-Numerous Utilities

The toolbox provides a series set of utilities which can help users to train the detection and recognition model.

Supported Algorithms:

Text Detection

• ToyDet
• Mask R-CNN (ICCV'2017)

Text Recognition

• CRNN (TPAMI'2016)
• ATTENTION ()

Layout Analysis

• CenterNet ()

Install

pip install requirements.txt

Training

sh shme.sh

Copyright

copyright