Tested for Ubuntu 18.04/20.04.
Use of a GPU significantly speeds up generation of detection outputs, but it is possible to run the inference demo code on CPU.
-
Set up python = 3.7.x environment:
pyenv install 3.7.12
pyenv virtualenv 3.7.12 dosa-env
-
Activate the environment
pyenv shell dosa-env
-
Update pip & setuptools
python -m pip install --upgrade pip setuptools
- Install requirements
pip install -r requirements.txt
- (for GPU-enabled installation:
pip install -r requirements_gpu.txt
)
- (for GPU-enabled installation:
-
Install Mask R-CNN
pip install -e ./Mask_RCNN
-
Install DocParser
pip install -e ./DocParser
- Download model weights follow instruction in
DocParser/docparser/default_models/README.md
- Insall PaddlePaddle
pip install paddlepaddle==2.1.3
- (for GPU-enabled installation:
pip install paddlepaddle-gpu==2.1.3
)
- (for GPU-enabled installation:
Installing paddlepaddle will raise warning error about dependency of gast==0.2.2 in tensorflow==1.15.5 vs. gast==0.4.0 in paddlepaddle==2.1.3. Just ignore it!
- Insall PaddleOCR
pip install -e ./PaddleOCR
-
Install poetry following instruction
https://github.com/python-poetry/poetry#osx--linux--bashonwindows-install-instructions
-
Install server dependencies
poetry install
Try each model with script in ./demos
, or running API server in ./server
and ./demos/server_api.py