This repository contains the code for reproducing the the paper:
Context-Based Contrastive Learning for Scene Text Recognition
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.
- 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
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
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
.
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
.
This project is based on the ABINet (Autonomous, Bidirectional and Iterative) codebase. Thank the authors for providing their implementation and making it publicly available.
- ConCLR: Paper Link
- ABINet: GitHub Repository