RK3588上测试paddle-fastdeploy-picodet样例（cpu可以推理，npu无法推理）

温馨提示：根据社区不完全统计，按照模板提问，可以加快回复和解决问题的速度

• 【FastDeploy版本】：git clone https://github.com/PaddlePaddle/FastDeploy.git

• 【编译命令】自行编译

• -安装步骤
  STEP1：
  git clone https://github.com/PaddlePaddle/FastDeploy.git
  cd FastDeploy
  mkdir build && cd build
  cmake .. -DENABLE_ORT_BACKEND=ON
  -DENABLE_RKNPU2_BACKEND=ON
  -DENABLE_VISION=ON
  -DRKNN2_TARGET_SOC=RK3588
  -DCMAKE_INSTALL_PREFIX=${PWD}/fastdeploy
  make -j8
  make install

STEP2：
永久配置
source /home/neardi/ruida/work/FastDeploy/build/fastdeploy/fastdeploy_init.sh
sudo cp /home/neardi/ruida/work/FastDeploy/build/fastdeployfastdeploy_libs.conf /etc/ld.so.conf.d/
sudo ldconfig

STEP3：
cd FastDeploy/examples/vision/detection/paddledetection/rknpu2/cpp/
mkdir build
cd build
cmake .. -DFASTDEPLOY_INSTALL_DIR=~/ruida/work/FastDeploy/build/fastdeploy
make -j8

• 【系统平台】: Debian11

• 【硬件】： RK3588

• 【编译语言】： C++

• 【模型跑不通】
  [CPU可以跑通，NPU跑不通]
  

• 【文件目录】

Start with examples/vision/detection/paddledetection/rknpu2/cpp/ and reproduce the build using the RK3588, Debian 11, and RKNPU2 CMake options listed in the issue. Compare the CPU and NPU execution output and inspect the referenced screenshots for the failure details. Done means the PicoDet example runs inference successfully on the NPU.