一、主机模型转换

采用FastDeploy来部署应用深度学习模型到OK3588板卡上

进入主机Ubuntu的虚拟环境
conda activate ok3588
主机环境搭建可以参考上一篇《OK3588板卡实现人像抠图（十二）》

转换成RKNN模型

cd FastDeploy
wget  https://bj.bcebos.com/paddlehub/fastdeploy/rknpu2/scrfd_500m_bnkps_shape640x640.zip
unzip scrfd_500m_bnkps_shape640x640.zip
python  tools/rknpu2/export.py \
        --config_path tools/rknpu2/config/scrfd_unquantized.yaml \
        --target_platform rk3588

得到scrfd_500m_bnkps_shape640x640_rk3588_unquantized.rknn 放到OK3588板卡上

二、板卡模型部署

进入虚拟环境
conda activate ok3588

git clone https://github.com/PaddlePaddle/FastDeploy.git
cd FastDeploy

# 如果您使用的是develop分支输入以下命令
git checkout develop

mkdir build && cd build
cmake ..  -DENABLE_ORT_BACKEND=OFF \
	      -DENABLE_RKNPU2_BACKEND=ON \
	      -DENABLE_VISION=ON \
	      -DRKNN2_TARGET_SOC=RK3588 \
          -DCMAKE_INSTALL_PREFIX=${PWD}/fastdeploy-0.0.0

make -j2
make install
#为了方便大家配置环境变量，FastDeploy提供了一键配置环境变量的脚本
source fastdeploy-0.0.0/fastdeploy_init.sh
sudo cp fastdeploy-0.0.0/fastdeploy_libs.conf /etc/ld.so.conf.d/
sudo ldconfig

cd FastDeploy/examples/vision/facedet/scrfd/rknpu2/cpp
mkdir build
cd build
cmake .. -DFASTDEPLOY_INSTALL_DIR=/home/forlinx/FastDeploy/build/fastdeploy-0.0.0/
make -j
得到了编译后的文件 infer_demo

三、执行推理

把scrfd_500m_bnkps_shape640x640_rk3588_unquantized.rknn放在build里面

NPU推理

找一张测试图片

wget https://raw.githubusercontent.com/DefTruth/lite.ai.toolkit/main/examples/lite/resources/test_lite_face_detector_3.jpg
./infer_demo scrfd_500m_bnkps_shape640x640_rk3588_unquantized.rknn test_lite_face_detector_3.jpg 1