djeong20 / text-recognizer

Text-Recognizer

Getting Started

Dependency

• requirements : numpy, pandas, matplotlib, pytorch, opencv-python, torchvision, albumentations, and Levenshtein

pip3 install numpy pandas tqdm matplotlib opencv-python torch torchvision albumentations Levenshtein

Download GNHK dataset for training from here

Folder structure

The structure of folder as below.

Text-Recognizer
├── README.md
├── dataset.py
├── model.py
├── train.py
├── utils.py
└── gnhk
    ├── test
    └── train

Check Data Pipeline

 python3 dataset.py --data='gnhk/train/' --transform=False 

Arguments

• --data: folder path to gnhk dataset.
• --transform: transformation. (default=False)

Training

 python3 train.py --train_data='gnhk/train/' --valid_data='gnhk/test/' \
 --train_trans=True --valid_trans=True \
 --batch_size=16 --n_epochs=5 --grad_clip=0.5 --workers=4

Arguments

• --train_data: folder path to gnhk training dataset.
• --valid_data: folder path to gnhk validation dataset.
• --train_trans: transformation in training. (default=True)
• --valid_trans: transformation in validation. (default=True)
• --batch_size: input batch size. (default=16)
• --n_epochs: number of epoches. (default=5)
• --grad_clip: gradient clipping value. (default=0.5)
• --workers: number of data loading workers. (default=4)