ChineseBrailleTranslation

📃 [Paper] • 💻 [Github] • 🤗 [Dataset] • ⚙️ [Model] • 🎬 [Demo]

Setup Environment

We recommend using a conda environment to manage dependencies.

conda create -n chinese_braille python=3.10
conda activate chinese_braille
pip install -r requirements.txt

The installation of pytorch may vary depending on your system. Please refer to the official website for more information.

All the training and evaluation scripts use accelerate to speed up the training process. If you want to run the scripts without accelerate, you can remove the accelerate related code in the scripts. Remember to run accelerate config before you run our scripts, or you may encounter some errors.

Data Preparation

Please follow the instructions in the data_preparation folder to prepare the dataset. We also provide three preprocessed dataset in HuggingFace:

• Violet-yo/Chinese-Braille-Dataset-10per-Tone
• Violet-yo/Chinese-Braille-Dataset-No-Tone
• Violet-yo/Chinese-Braille-Dataset-Full-Tone

All these three dataset are used in our training.

Training

Add Special Tokens

First we need to add the braille characters to the tokenizer. We provide a script to add the special tokens to the tokenizer. You can run the script by:

python mt5_add_special_tokens.py --original_model_dir $T5_ORIGINAL_DIR --output_dir $T5_SPECIAL_DIR

where $T5_ORIGINAL_DIR is the directory of the original T5 model and $T5_SPECIAL_DIR is the directory to save the new model with special tokens and extended word embedding weights.

Fine Tuning

We provide an example script in run_translation_accelerate.sh to train the model. You can modify the script to fit your needs. Note that we use three stages to train the model, as stated in the paper. The training dataset and training arguments may need to be changed for each stage.

We provide a fine-tuned checkpoint in HuggingFace:

• Violet-yo/mt5-small-ft-Chinese-Braille

Evaluation

Test Inference

We provide an example inference script. You can run it by:

python test_inference_simp.py

The script will load the fine-tuned model and translate the input text to braille. You can modify the input text in the script.

Evaluate on the Validation and Test Set

You can get the score on the validation and test set by running:

bash run_translation_evaluation.sh

Contact

If you have any questions about out project, please feel free to emial to ayw34@cornell.edu.

Citation

@misc{wu2024visionbrailleendtoendtoolchinese,
      title={Vision-Braille: An End-to-End Tool for Chinese Braille Image-to-Text Translation}, 
      author={Alan Wu and Ye Yuan and Ming Zhang},
      year={2024},
      eprint={2407.06048},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2407.06048}, 
}