I'm trying to use SNPE to run Deep learning models on the Snapdragon Flight board. I'm having difficulties using "snpe-net-run" on the Snapdragon Flight board.
To be specific, I'm stuck in the "Run on android target" part of the tutorial for the "Tensorflow inception v3" model. Once I push all the binaries, I'm unable to run "snpe-net-run -h" on the target (via adb shell). I think this could be an issue with the android binaries' architecture. I tried all four combinations listed: armeabi-v7a and arm64-v8a, one compiled with gcc4.9 using gnustl STL implementation, and the other compiled with clang3.8 using libc++ STL implementation.
Has anyone done this? I'm not trying to execute a custom model, just the inbuilt tutorial ones.
My version of SNPE is snpe-1.12.0.
I'm new to SNPE and the Snapdragon flight board, so apologies if I've left out important information. I'm willing to provide info if necessary.
Thank you.
Resolved by doing the following:
Copying the following SNPE binaries to the board:
snpe-1.12.0/bin/arm-linux-gcc4.8hf/snpe-net-run
snpe-1.12.0/lib/arm-linux-gcc4.8hf/*.so
On the board, change directory to where the files are copied.
chmod u+x snpe-net-run
export LD_LIBRARY_PATH=${PWD}
snpe-net-run should run. Need to be root to use GPU acceleration, by “sudo su”.