OCR_resnet

Project Description

This project is a Optical Character Recognition task.
This task is to recognize Korean from a selected dataset by useing EAST model.

File Description

• model.py

  code for constructing EAST model

• loss.py

code for defining the loss function for training

• east_dataset.py

  code for construcing a dataset class for training model

• dataset.py

  code for constructing a dataset class that provides images and text bounding boxes

• train.py

  code for implementing model training

• inference.py

  code for conducting model inference process

• detect.py

  code for constructing inference and detection functions

• deteval.py

  code for constructing functions for DetEval

• convert_mlt.py

  code for transfer ICDAR dataset images into ufo format

• requirements.txt

  requirements for implementation

• /eval_dataset

  folder for evaluation image dataset

• /pth

  folder for storing model pretraining files

• /predictions

  folder for storing inference results file

Dataset Description

This project uses ICDAR-2017 dataset
Download Link:ICDAR-2017

How to train?

• Step 1

  Download dataset from the link above and put all images into one folder, then implement convert_mlt.py to tranfer the images infomation into ufo format.

• Step 2

  pip install -r requirements.txt to install the required python libraries

• Step 3

  Training: python train.py
  Inference: python inference.py