Build for Android ARMv7

[marimomo00]

Thanks for the sample project. It helps me a lot.

I'm new to building App on Unity. I'm really need help from you. I'm trying to build the sample of Posenet for Android. I can build the Android App with ARM64, it works very well on my mobile. However, it doesn't work on my another android phone. So I tried to build with ARMv7 support. It built successfully but it runs with error. The error is,

i>AndroidPlayer(ADB@127.0.0.1:34999)</i DllNotFoundException: Unable to load DLL 'libtensorflowlite_gpu_delegate': The specified module could not be found. at TensorFlowLite.GpuDelegate.TfLiteGpuDelegateOptionsV2Default () [0x00000] in <00000000000000000000000000000000>:0 at TensorFlowLite.GpuDelegate..ctor () [0x00000] in <00000000000000000000000000000000>:0 at TensorFlowLite.BaseImagePredictor1[T].CreateGpuDelegate () [0x00000] in <00000000000000000000000000000000>:0 at TensorFlowLite.BaseImagePredictor1[T]..ctor (System.String modelPath, System.Boolean useGPU) [0x00000] in <00000000000000000000000000000000>:0 at TensorFlowLite.PoseNet..ctor (System.String modelPath) [0x00000] in <00000000000000000000000000000000>:0 at PoseNetSample.Start () [0x00000] in <00000000000000000000000000000000>:0

What is the problem? Anything that I can do?

[asus4]

@marimomo00 Android plugin *.so files are built for only Arm64 CPU for now. Could you try to rebuild the plugin with android_arm option instead of android_arm64?

# Configure the Android SDK path by running configure script at the repository root
./configure

# Build experimental
bazel build -c opt --cxxopt=--std=c++11 --config=android_arm //tensorflow/lite/experimental/c:libtensorflowlite_c.so

# Build GPU delegate
bazel build -c opt --config android_arm --copt -Os --copt -DTFLITE_GPU_BINARY_RELEASE --copt -fvisibility=hidden --linkopt -s --strip always //tensorflow/lite/delegates/gpu:libtensorflowlite_gpu_delegate.so

[marimomo00]

Thanks for your prompt reply.

I will have a try.

On Mon, Apr 6, 2020 at 6:46 PM Koki Ibukuro notifications@github.com wrote:

@marimomo00 https://github.com/marimomo00 Android plugin *.so files are built for Arm64 CPU for now. Could you try to rebuild the plugin with android_arm option instead of android_arm64?

Configure the Android SDK path by running configure script at the repository root

./configure

Build experimental

bazel build -c opt --cxxopt=--std=c++11 --config=android_arm //tensorflow/lite/experimental/c:libtensorflowlite_c.so

Build GPU delegate

bazel build -c opt --config android_arm --copt -Os --copt -DTFLITE_GPU_BINARY_RELEASE --copt -fvisibility=hidden --linkopt -s --strip always //tensorflow/lite/delegates/gpu:libtensorflowlite_gpu_delegate.so

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/asus4/tf-lite-unity-sample/issues/11#issuecomment-609719138, or unsubscribe https://github.com/notifications/unsubscribe-auth/APCOC73JOZ27CRVMMIMDIJTRLGXH7ANCNFSM4MCEXEDQ .

[gblue1223]

I wanna share my experience about Bazel building. First of all, I've built with ''tensorflow-2.1.0" and ''bazel-0.29.1''. And I ran ''./configure'' and everything leave as default but set options regarding Android.

Then I ran the command as @asus4 said. You have to make sure about tensorflow and bazel version. It is very sensitive.

[manena]

I'm sorry I'm new to building so files. I would like to build for android_arm. I'm working on MAC OS Catalina and have no idea if these are commands to run from the shell, what directory I should be in and what would I need to install in order to be able to build.

Can you give me some guidance?

Thanks in advance

[marimomo00]

@manena There are instruction on building tensorflow. https://www.tensorflow.org/install/source

However, I got lots of error when compiling on windows. I'm still struggling on building android_arm. I've opened another issue and hope someone could help.

https://github.com/tensorflow/tensorflow/issues/38525

@gblue1223 I've tried ''tensorflow-2.1.0" and ''bazel-0.29.1'', but with error. Do you build on Windows or MAC OS?

[gblue1223]

@marimomo00 I've built on Ubuntu. Since Tensorflow more familiar with Linux, you should try on Linux OS.

[manena]

Thanks @marimomo00 i’ll try. In the meanwhile @gblue1223: would it be possible for you to share with us the .so files you have built? It would be very helpful!

Regards and I hope all of you and your families are fine!

[gblue1223]

@manena Here is the .so files. Good luck~

[Freeverc]

I've built libtensorflowlite_c.so for ARMv7，but I can not use it in your samples. The result on my phone is pure white with no images.How can I use it in ARMv7? I've changed the Target Architectures.

[emusiceducation123]

I built for ARMv7 as well. When it runs, log shows: 08-29 11:01:18.646 3872 4085 I tflite : Initialized TensorFlow Lite runtime. 08-29 11:01:20.242 7703 7723 E Unity : Unable to find libtensorflowlite_gpu_delegate 08-29 11:01:20.244 7703 7723 E Unity : Unable to find tensorflowlite_gpu_delegate 08-29 11:01:20.259 7703 7723 E Unity : Unable to find libtensorflowlite_c

[mathtasatlime]

Does anybody have a working build auf 2.4.0 for armv7 ?

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.