- Linux (Windows is not officially supported now)
- Python 3.7
- Pytorch >= 1.6
- torchvision 0.7.0
- CUDA 10.1
Our version OS and packages versions we used for bellow tests was:
- OS: Ubuntu 18.04
- CUDA: 10.1
- GCC (G++): 8.4.0 (CUDA 10.1 do not support for g++ version later 8)
- Pytorch: 1.6.0
- torchvision: 0.7.0
- Create conda virtual environment and afterward activate it.
conda create -n gen_ocr python=3.7 -y
conda activate gen_ocr
- Install Pytorch and torchvision. following the official Pytorch instructions
conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch
Make sure that your CUDA compilation and CUDA runtime is matched version in order to avoid unexpected error. To check your CUDA runtime version in termial `nvcc --version`.
- Clone general_ocr repository
git clone https://github.com/phamdinhkhanh/general_ocr.git
cd general_ocr
- Download
onnxruntime-linux
from ONNX Runtime releases, extract it, exposeONNXRUNTIME_DIR
and finally add the lib path toLD_LIBRARY_PATH
as below:
wget https://github.com/microsoft/onnxruntime/releases/download/v1.8.1/onnxruntime-linux-x64-1.8.1.tgz
tar -zxvf onnxruntime-linux-x64-1.8.1.tgz
cd onnxruntime-linux-x64-1.8.1
export ONNXRUNTIME_DIR=$(pwd)
export LD_LIBRARY_PATH=$ONNXRUNTIME_DIR/lib:$LD_LIBRARY_PATH
- Install
onnxruntime
pip install onnxruntime==1.8.1
- Install build requirements packages.
cd .. # general_ocr root repository directory
pip install -r requirements.txt
GENERAL_OCR_WITH_OPS=1 GENERAL_OCR_WITH_ORT=1 python -m pip install -v -e .
export PYTHONPATH=$(pwd):$PYTHONPATH
Full setup script is there:
# create and activate conda enviroment
conda create -n gen_ocr python=3.7 -y
conda activate gen_ocr
# install pytorch and torchvision
conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch
# clone general_ocr
git clone https://github.com/phamdinhkhanh/general_ocr.git
cd general_ocr
# download onnxruntime-linux and expose its lib to LD_LIBRARY_PATH
wget https://github.com/microsoft/onnxruntime/releases/download/v1.8.1/onnxruntime-linux-x64-1.8.1.tgz
tar -zxvf onnxruntime-linux-x64-1.8.1.tgz
cd onnxruntime-linux-x64-1.8.1
export ONNXRUNTIME_DIR=$(pwd)
export LD_LIBRARY_PATH=$ONNXRUNTIME_DIR/lib:$LD_LIBRARY_PATH
# install onnxruntime
pip install onnxruntime==1.8.1
# install requirements.txt
cd .. # general_ocr root repository directory
pip install -r requirements.txt
GENERAL_OCR_WITH_OPS=1 GENERAL_OCR_WITH_ORT=1 python -m pip install -v -e .
export PYTHONPATH=$(pwd):$PYTHONPATH
Run script below to test your installation is successful:
conda activate gen_ocr
python general_ocr/utils/ocr.py demo/demo_text_ocr.jpg --print-result --imshow --det PANet_IC15 --recog SEG