This repo combine the power of heat map based text detection method with advance backbone from efficient line-up.
├── detec
│ ├── efficient
│ └── craft
└── recog
└── atten
Install pytorch:
pip3 install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html
Use requirements.txt (note: use pip/pip3 and python/python3 base on your environment settings)
cd akaocr/
pip3 install -r requirements.txt
If the installation get error or cannot found packages, you may need to upgrade your pip first:
pip3 install --upgrade pip
Run bash script:
cd akaocr/dataprep
sh prepare_detec.sh
Replace (akaocr) with relative path and match the target train/test set of dataset in akaocr/data. For ex, for ICDAR13:
--data_detec=(akaocr)/data/icdar13/train
--data_test=(akaocr)/data/icdar13/test
Train detec:
cd akaocr/tools
python3 train_detec.py --data_detec=../data/<dataset_name>/train --data_test_detec=../data/<dataset_name>/test --exp=<experiment_name> --weight=<pretrain_model>
List of dependencies
torch==1.5.1
torchvision==0.6.1
opencv-python==4.4.0.42
tqdm==4.43.0
h5py==2.10.0
imageio==2.8.0
imutils==0.5.3
lmdb==0.98
natsort==7.0.1
nltk==3.5
numpy==1.18.1
pandas==1.0.1
pdf2image==1.13.1
pillow==7.1.2
polygon3==3.0.8
pydot==1.4.1
scipy==1.4.1
scikit-image==0.16.2
tabulate==0.8.6
pygame==2.0.0.dev10
imagecorruptions==1.1.0
imageio==2.8.0
imgaug==0.4.0
MIT License
Email - nguyenvietbac1@gmail.com
Email - nnnghia.96@gmail.com
Email - huukim98@gmail.com