PaddleOCR转RKNN并在RK3588上运行

一、Ubuntu20.04（PC端）

1.创建环境

conda create -n deploy python=3.6 conda activate deploy

2.安装环境

链接: https://pan.baidu.com/s/18KoIYMkCUS5-PNb3DCRTHA?pwd=5ix5 提取码: 5ix5

pip install rknn_toolkit2-1.4.2b3+0bdd72ff-cp36-cp36m-linux_x86_64.whl pip install paddle2onnx

3.PaddleOCR转换

3.1 下载PP-OCRv3文字检测模型、文字方向分类器模型、PP-OCRv3文字识别模型

wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar tar -xvf ch_PP-OCRv3_det_infer.tar wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar tar -xvf ch_ppocr_mobile_v2.0_cls_infer.tar wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar tar -xvf ch_PP-OCRv3_rec_infer.tar

3.2 转换成ONNX，并将ONNX固定输入shape

paddle2onnx --model_dir ch_PP-OCRv3_det_infer \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --opset_version 11 \ --save_file det3.onnx \ --enable_dev_version True paddle2onnx --model_dir ch_ppocr_mobile_v2.0_cls_infer \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --opset_version 11 \ --save_file cls2.onnx \ --enable_dev_version True paddle2onnx --model_dir ch_PP-OCRv3_rec_infer \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --opset_version 11 \ --save_file rec3.onnx \ --enable_dev_version True python -m paddle2onnx.optimize --input_model det3.onnx \ --output_model det3_shape.onnx \ --input_shape_dict "{'x':[1,3,960,960]}" python -m paddle2onnx.optimize --input_model cls2.onnx \ --output_model cls2_shape.onnx \ --input_shape_dict "{'x':[1,3,48,192]}" python -m paddle2onnx.optimize --input_model rec3.onnx \ --output_model rec3_shape.onnx \ --input_shape_dict "{'x':[1,3,48,320]}"

3.3 转换成RKNN

# 下载FastDeploy1.1.0版本，除了可以python调用也支持C++ # 下载解压后先修改该目录下rknpu2_tools/config中三个yaml，更换onnx路径和rknn输出路径后开始转换 git clone https://github.com/PaddlePaddle/FastDeploy.git cd FastDeploy-release-1.1.0/examples/vision/ocr/PP-OCR/rockchip/rknpu2_tools python rknpu2_tools/export.py --config_path rknpu2_tools/config/ppocrv3_det.yaml --target_platform rk3588 python rknpu2_tools/export.py --config_path rknpu2_tools/config/ppocrv3_rec.yaml --target_platform rk3588 python rknpu2_tools/export.py --config_path rknpu2_tools/config/ppocrv3_cls.yaml --target_platform rk3588

得到三个RKNN文件，拷贝至RK3588版端

二、RK3588板端

1.搭建环境

先下载rknn-toolkit-lite2

链接: https://pan.baidu.com/s/1mb8XmoTSVz2ZsO9HnKvr8Q?pwd=5ax6 提取码: 5ax6

conda create -n deploy python=3.8 conda activate fastdeploy pip install packages/rknn_toolkit_lite2-2.0.0b1-cp38-cp38-linux_aarch64.whl