/paddleocr.cpp

PaddleOCR + OnnxRuntime

Primary LanguageC++

Just clone PaddleOCR cpp_infer, and using OnnxRuntime instead of Paddle to run it on Mac M1

1. Export the inference model

Download PaddleOCR ocr models and layout models, and directory structure is as follows

inference/
|-- det_db
|   |--inference.pdiparams
|   |--inference.pdmodel
|-- rec_rcnn
|   |--inference.pdiparams
|   |--inference.pdmodel
|-- cls
|   |--inference.pdiparams
|   |--inference.pdmodel
|-- table
|   |--inference.pdiparams
|   |--inference.pdmodel
|-- layout
|   |--inference.pdiparams
|   |--inference.pdmodel

Export to Onnx using paddle2onnx.

paddle2onnx --model_dir saved_inference_model \
            --model_filename inference.pdmodel \
            --params_filename inference.pdiparams\
            --save_file inference.onnx \
            --enable_dev_version True

Onnx model structure.

inference/
|-- det_db
|   |--inference.onnx
|-- rec_rcnn
|   |--inference.onnx
|-- cls
|   |--inference.onnx
|-- table
|   |--inference.onnx
|-- layout
|   |--inference.onnx

2. Compile PaddleOCR

mkdir build
cd build
cmake ..
make -j

Note: still in cmake dev to build dependencies as static libs

3. Run the demo

1. det+cls+rec：
./build/PaddleOcrOnnx 
    --det_model_dir=inference/det_db \
    --rec_model_dir=inference/rec_rcnn \
    --cls_model_dir=inference/cls \
    --image_dir=images/1.jpg \
    --use_angle_cls=true \
    --det=true \
    --rec=true \
    --cls=true \
2. layout
./build/PaddleOcrOnnx 
    --layout_model_dir=inference/layout \
    --image_dir=images/1.png \
    --type=structure \
    --table=false \
    --layout=true \
    --det=false \
    --rec=false
3. table
./build/PaddleOcrOnnx 
    --det_model_dir=inference/det_db \
    --rec_model_dir=inference/rec_rcnn \
    --table_model_dir=inference/table \
    --image_dir=images/table.jpg \
    --type=structure \
    --table=true

4. Reference

PaddleOCR cpp_infer: origin implementation of PaddleOCR cpp

PaddleOCR + OnnxRuntime: the repo has no layout function