PP-OCRv3 was proposed by baidu, it contains a series of tasks with multiple models, including
- Text detection
DBDetector - [Optional] Direction classification
Classiferis used to adjust the direction of images before text recognition - Character recognition
Recognizeris used to recognize characters from images
Paddle2ONNX needs to be installed for the conversion of model
#1. change the Paddle model from dynamic inputs to fixed inputs, see paddle_infer_shape.py
#det model
python3 paddle_infer_shape.py --model_dir ch_PP-OCRv3_det_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_dir ch_PP-OCRv3_det_infer \
--input_shape_dict="{'x':[1,3,960,608]}"
#cls model
python3 paddle_infer_shape.py --model_dir ch_ppocr_mobile_v2.0_cls_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_dir ch_ppocr_mobile_v2.0_cls_infer \
--input_shape_dict="{'x':[1,3,48,192]}"
#rec model
python3 paddle_infer_shape.py --model_dir ch_PP-OCRv3_rec_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_dir ch_PP-OCRv3_rec_infer \
--input_shape_dict="{'x':[1,3,48,584]}"
#2. Convert the Paddle model with fixed inputs to the ONNX model, see paddle2onnx
#det model
paddle2onnx --model_dir ch_PP-OCRv3_det_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_file ch_PP-OCRv3_det_infer_fix.onnx \
--enable_dev_version True
#cls model
paddle2onnx --model_dir ch_ppocr_mobile_v2.0_cls_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_file ch_ppocr_mobile_v2.0_cls_infer.onnx \
--enable_dev_version True
#rec model
paddle2onnx --model_dir ch_PP-OCRv3_rec_infer \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--save_file ch_PP-OCRv3_rec_infer.onnx \
--enable_dev_version True| Model | Download |
|---|---|
| ch_PP-OCRv3_det | 3.6 MB |
| ch_ppocr_mobile_v2.0_cls | 2.1 MB |
| ch_PP-OCRv3_rec | 11.3 MB |
Apache 2.0