当前位置：首页 > news >正文

高通SNPE测试：6、在开发板上运行Inception v3 Model（oe-linux）

news 2025/8/26 9:45:31

实际上还是很顺利的，终于来到了使用开发板进行测试了。

环境变量

教程是aarch64-android，我的板子是linux。

# For Android targets: architecture: arm64-v8a - compiler: clang - STL: libc++
# export SNPE_TARGET_ARCH=aarch64-android
# Example for LE targets
export SNPE_TARGET_ARCH=aarch64-oe-linux-gcc11.2

复制文件1：adb shell方式

# For Android targets: architecture: arm64-v8a - compiler: clang - STL: libc++
# export SNPE_TARGET_ARCH=aarch64-android
# Example for LE targets
export SNPE_TARGET_ARCH=aarch64-oe-linux-gcc11.2export SNPE_TARGET_DSPARCH=hexagon-v73DST_DIR=${DST_ROOT}/data/local/snpe/snpeexampleadb shell "mkdir -p ${DST_DIR}/$SNPE_TARGET_ARCH/bin"
adb shell "mkdir -p ${DST_DIR}/$SNPE_TARGET_ARCH/lib"
adb shell "mkdir -p ${DST_DIR}/dsp/lib"adb push $SNPE_ROOT/lib/$SNPE_TARGET_ARCH/*.so \${DST_DIR}/$SNPE_TARGET_ARCH/lib
adb push $SNPE_ROOT/lib/$SNPE_TARGET_DSPARCH/unsigned/*.so \${DST_DIR}/dsp/lib
adb push $SNPE_ROOT/bin/$SNPE_TARGET_ARCH/snpe-net-run \${DST_DIR}/$SNPE_TARGET_ARCH/bincd $SNPE_ROOT/examples/Models/InceptionV3
mkdir data/rawfiles && cp data/cropped/*.raw data/rawfiles/DST_DIR=/data/local/snpe/inception_v3
adb shell "mkdir -p                       ${DST_DIR}/cropped"
adb push   data/rawfiles                  ${DST_DIR}/cropped
adb push   data/target_raw_list.txt       ${DST_DIR}
adb push   dlc/inception_v3_quantized.dlc ${DST_DIR}
rm -rf data/rawfiles

复制文件2：tar方式

比如我是虚拟机，先打包，再复制到开发板上解压移动过去。

# For Android targets: architecture: arm64-v8a - compiler: clang - STL: libc++
# export SNPE_TARGET_ARCH=aarch64-android
# Example for LE targets
export SNPE_TARGET_ARCH=aarch64-oe-linux-gcc11.2export SNPE_TARGET_DSPARCH=hexagon-v73WORK_PATH=`pwd`PLATFORM_NAME=sg882g
DST_ROOT=${WORK_PATH}/${PLATFORM_NAME}
if [ -d ${DST_ROOT} ]; thenrm -rf ${DST_ROOT}
fi
SNPE_ROOT_PATH=/data/local/snpe
DST_DIR=${DST_ROOT}/${SNPE_ROOT_PATH}/snpeexamplemkdir -p ${DST_DIR}/$SNPE_TARGET_ARCH/bin
mkdir -p ${DST_DIR}/$SNPE_TARGET_ARCH/lib
mkdir -p ${DST_DIR}/dsp/libcp  -rfP $SNPE_ROOT/lib/$SNPE_TARGET_ARCH/*.so \${DST_DIR}/$SNPE_TARGET_ARCH/lib
cp  -rfP $SNPE_ROOT/lib/$SNPE_TARGET_DSPARCH/unsigned/*.so \${DST_DIR}/dsp/lib
cp  -rfP $SNPE_ROOT/bin/$SNPE_TARGET_ARCH/snpe-net-run \${DST_DIR}/$SNPE_TARGET_ARCH/binDST_DIR=${DST_ROOT}/${SNPE_ROOT_PATH}/inception_v3
cd $SNPE_ROOT/examples/Models/InceptionV3
mkdir -p ${DST_DIR}/croppedcp -rfP data/cropped/*.raw             ${DST_DIR}/cropped
cp -rfP data/target_raw_list.txt       ${DST_DIR}
cp -rfP dlc/inception_v3_quantized.dlc ${DST_DIR}cd ${WORK_PATH}cp copy-to-sg882g.sh ${PLATFORM_NAME}
cp  run-on-sg882g.sh ${PLATFORM_NAME}TAR_FILE=${PLATFORM_NAME}.tar
if [ -f ${TAR_FILE} ]; thenrm ${TAR_FILE}
fi
tar -czf ${TAR_FILE} ${PLATFORM_NAME}DST_DIR=${HOME}/share/E
if [ -f ${DST_DIR}/${TAR_FILE} ]; thenrm  ${DST_DIR}/${TAR_FILE}
fi
cp ${TAR_FILE} ${DST_DIR}

其中的两个shell，一个是复制文件到系统，一个是运行。

运行

注意，执行文件复制到/bin，库文件复制到/lib

# For Android targets: architecture: arm64-v8a - compiler: clang - STL: libc++
# export SNPE_TARGET_ARCH=aarch64-android
# Example for LE targets
export SNPE_TARGET_ARCH=aarch64-oe-linux-gcc11.2SNPE_ROOT_PATH=/data/local/snpe
SNPE_ARCH_PATH=${SNPE_ROOT_PATH}/snpeexample/$SNPE_TARGET_ARCHexport LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${SNPE_ARCH_PATH}/lib
export PATH=$PATH:${SNPE_ARCH_PATH}/bincd ${SNPE_ROOT_PATH}/inception_v3snpe-net-run \--container inception_v3_quantized.dlc \--input_list target_raw_list.txt-------------------------------------------------------------------------------
Model String: N/A
SNPE v2.21.0.240401120655_85612
-------------------------------------------------------------------------------
Processing DNN input(s):
cropped/notice_sign.raw
Processing DNN input(s):
cropped/trash_bin.raw
Processing DNN input(s):
cropped/chairs.raw
Processing DNN input(s):
cropped/plastic_cup.raw
Successfully executed!

运行结果

具体输出在output目录下，可以使用adb导出。

如果是在windows上，最好用tar打包出来。

adb pull /data/local/snpe/inception_v3/output output_linux

检查运行结果

cd snpe-2.21.0/examples/Models/InceptionV3python3 scripts/show_inceptionv3_classifications_snpe.py \-i data/target_raw_list.txt \-o output_linux/ \-l data/imagenet_slim_labels.txtcropped/notice_sign.raw 0.140213 459 brass
cropped/trash_bin.raw   0.772629 413 ashcan
cropped/chairs.raw      0.301875 832 studio couch
cropped/plastic_cup.raw 0.976873 648 measuring cup

对比在UBUNTU上的结果，有所差异。

查看全文

http://www.dtcms.com/a/350328.html