This is the official implementation of our Pattern Recognition(PR) 2025 paper "HTR-VT: Handwritten Text Recognition with Vision Transformer". It's a new and effective baseline for handwritten text recognition solely using Vision Transformer and CTC Loss.

[Project Page] [Paper] [arXiv] [Datasets] [Checkpoints]

• 1. Overview
• 2. Visual Results
• 3. Installation
• 4. Quick Start
• 5. Citation
• 6. Acknowledgement

Our model can be learnt in a single GPU RTX-4090 24G

conda env create -f environment.yaml
conda activate htr

The code was tested on Python 3.9 and PyTorch 1.13.0.

• Using IAM, READ2016 and LAM for handwritten text recognition.

Register at the FKI's webpage :https://fki.tic.heia-fr.ch/databases/iam-handwriting-database)
Download the dataset from here :https://fki.tic.heia-fr.ch/databases/download-the-iam-handwriting-database

wget https://zenodo.org/record/1164045/files/{Test-ICFHR-2016.tgz,Train-And-Val-ICFHR-2016.tgz}

Download the dataset from here: https://aimagelab.ing.unimore.it/imagelab/page.asp?IdPage=46

• Download datasets to ./data/. Take IAM for an example: The structure of the file should be:

./data/iam/
├── train.ln
├── val.ln
├── test.ln
└── lines
      ├──a01-000u-00.png
      ├──a01-000u-00.txt
      ├──a01-000u-01.png
      ├──a01-000u-01.txt
      ...

• We provide convenient and comprehensive commands in ./run/ to train and test on different datasets to help researchers reproducing the results of the paper.

If our project is helpful for your research, please consider citing :

@article{li2025htr,
  title={HTR-VT: Handwritten text recognition with vision transformer},
  author={Li, Yuting and Chen, Dexiong and Tang, Tinglong and Shen, Xi},
  journal={Pattern Recognition},
  volume={158},
  pages={110967},
  year={2025},
  publisher={Elsevier}
}

We appreciate helps from public code: VAN and OrigamiNet.