Recognizing tables with Chinese characters and numbers in handwriting tickets
Run the following commands to install OpenVino.
cd $project_root
pip install openvino-dev
omz_downloader --name handwritten-simplified-chinese-recognition-0001
If calling omz_downloader
fails, it is at $venv/lib/site-packages/openvino/model_zoo/omz_downloader.py
,
where $venv
is the Python virtual environment directory.
Anaconda
Copy Python library into OpenVino. The location of Python library is different depending on operating systems. Generally, if you use Anaconda in Ubuntu and your environment is named
ocr
, Python library is at~/.conda/envs/ocr/lib/libpython3.9.so.1.0
and OpenVino is at~/.conda/envs/ocr/lib/python3.9/site-packages/openvino/libs
.
Follow the guidance of https://pytorch.org/get-started/locally/ to install PyTorch. If you don't have an Intel GPU, please install CPU version.
Run the following commands to install other packages. Remove PyTorch in this file if you have installed it and configured carefully.
pip install -r requirements.txt
Intel GPU
If you have a Intel GPU, please choose the calculation component that is compatible to both OpenVino and PyTorch. Also in
bulk_recognition.ipynb
, please replace the following linenet_exec = ie.load_network(network=net, device_name='CPU')
with
net_exec = ie.load_network(network=net, device_name='GPU')
This config hasn't been tested because the author does not have a Intel GPU.
This project use the compiled binary form of LSTM-RNN-CTC pretrained model by Intel when testing, but does not have right to take actions including but not limiting to redistribute, copy, modify the upstreaming resources.
When you have a bulk of tickets, which contains similar table which in a similar
position in each paper, please combine the images into a PDF file, where pages
have the same size, background color is light, and text color is dark. Then, follow
bulk_recognition.ipynb
to analyze it.