ConCLR: Context-Based Contrastive Learning for Scene Text Recognition

This repository contains the code for reproducing the the paper:

Context-Based Contrastive Learning for Scene Text Recognition

Introduction

This project aims to reproduce the results of the paper "Context-Based Contrastive Learning for Scene Text Recognition". The main contribution of this paper is the introduction of a context-based contrastive learning framework to improve the robustness and accuracy of scene text recognition models.

Requirements

• Python 3.7 or higher
• PyTorch 1.1.0 or higher
• TensorBoard (optional, for monitoring training progress)
• Other dependencies listed in requirements.txt

Install the required packages using:

pip install -r requirements.txt

Project Structure

The project directory is structured as follows:

ConCLR/
├── data/                   # Dataset directory
├── models/                 # Model definitions
├── checkpoints/            # Model checkpoints
├── notebooks/              # Jupyter notebooks
├── configs/                # Configuration file directory
├── runs/                   # Traning logs
├── README.md               # This file
├── requirements.txt        # Required packages
└── *.py                    # Python scripts

Training

To train the model, run the following command:

python train.py --config configs/config.yaml

This will start the training process using the configuration specified in config.yaml.

TensorBoard

You can monitor the training process using TensorBoard. Run the following command to start TensorBoard:

tensorboard --logdir=runs/{experiment_name}

Then, open your web browser and go to http://localhost:6006.

Acknowledgements

This project is based on the ABINet (Autonomous, Bidirectional and Iterative) codebase. Thank the authors for providing their implementation and making it publicly available.

References

• ConCLR: Paper Link
• ABINet: GitHub Repository