Hi @luispimo
I have already installed the deployment package in the Pumpkin board successfully.
However, when I tests the IMGDNN Unit (imgdnn_test) under the ncsdk/gpu_aarch64/bin folder, some errors occurs. Although the testing finally passed, I wonder if the error message is expected?
Best regards,
Hight
Hi @hight
The goal of these tests is to verify that errors are correctly detected.
If you see the word "PASSED" everything is supposed to be OK.
Regards,
Luis
Hi @luispimo
I am running NCSDK on Raspbery PI 4. So far I could install Ubuntu, OpenCL, Caffe, Python after changing few lines in the Makefile.Config. It could successfully compile and run all Caffe codes.
Now I am at stage to perform imgdnn_test. I am having an environment variable problem as follows;
memi@memi-desktop:~/pumpkin$ sudo gpu_aarch64/bin/imgdnn_test
[sudo] password for memi:
Running main() from /usr/buildxl/buildroot/nc_sdk_src/bscontrol/obj/aarch64_binary-prefix/src/aarch64_binary-build/imgdnn/test/googletest-src/googletest/src/gtest_main.cc
[==========] Running 489 tests from 211 test suites.
[----------] Global test environment set-up.
/usr/buildxl/buildroot/nc_sdk_src/gpu/imgdnn/test/imgdnn_test.cpp:84: Failure
Expected: (nullptr) != (device), actual: (nullptr) vs NULL
[----------] Global test environment tear-down
Segmentation fault
memi@memi-desktop:~/pumpkin$
Could please advise what should I change in which file?
Thank you.
Dear Memi
You did well to get to the stage you did. But....
The function of our Neural Compute SDK software is to map from TensorFlow and Caffe on to Imagination architectures such as our PowerVR GPUs and NNx Neural Network Accelerators.
There is no GPU on an Raspberry Pi, so it will not work. The manuals specify the Ubuntu versions and explicitly mention x64 desktop architecture and GPU.
Also, the RPi family of boards do not contain any Imagination IP, so our motivation to get involved in these boards is minimal.
In addition to the SEEED Studio Pumpkin board, our PowerVR 8XE GPU is now accessible in the new BeagleBone (tm) AI-64 Board, which is on sale now.
It is well-supported and widely distributed so it will be the chosen platform for the public release of these EdgeAI teaching materials.
Thanks for your continued activity. It is all helping to improve these materials!
Best Regards,
Robert Owen