Conformer OCR is an Optical Character Recognition toolkit built for researchers working on both OCR for both Vietnamese and English. This project only focused on variants of vanilla Transformer and Feature Extraction (CNN-based approach).
This is also the first repo to utilize ConformerNet (https://arxiv.org/abs/2005.08100) for OCR.
- Variants of Transformer (e.g., Vanilla, Conformer) encoder with CTC decoder.
- Both naive Pytorch and Pytorch Lightning are provided
- Beam search with N-gram Language model
- Accumulation gradient training
cd transformer_ocr
pip install -r requirements/requirements.txt
To modulize the repo, the current structure is adopted as follows:
├── conf # configurations
│ ├── dataset
│ ├── model
│ ├── optimizer
│ ├── pl_params
│ └── config.yaml
├── requirements # Where store different requirements if needed
│ └── requirements.txt
├── scripts # Where start your training/evaluation/testing models
│ ├── train.py
│ └── train_PT.py
├── transformer_ocr # Main resource
└── README.md Train with naive Pytorch mode
cd scripts
python train.py
Train with Pytorch Lightning mode
cd scripts
python train_PT.py
Coming soon...