tensorflow: Android tensorflow-lite-select-tf-ops.aar generated from tflite model build and deploy error

Please make sure that this is a build/installation issue. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:build_template

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 18.04
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device:
• TensorFlow installed from (source or binary):
• TensorFlow version:
• Python version: 3.6
• Installed using virtualenv? pip? conda?:
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:

Describe the problem

I am following this guide to reduce my apk size. I followed the Set up build environment using Docker process to generate, tensorflow-lite.aar, tensorflow-lite-select-tf-ops.aar with my tflite model.

After running, bash tensorflow/lite/tools/build_aar.sh --input_models=custom_model.tflite --target_archs=arm64-v8a,armeabi-v7a I found the aar files in tmp folder of docker container in /tensorflow_src/bazel-bin/tmp/tensorflow-lite.aar and /tensorflow_src/bazel-bin/tmp/tensorflow-lite-select-tf-ops.aar.

I followed the part below by making libs folder and copying the aar files.

allprojects {
    repositories {
        jcenter()
        flatDir {
            dirs 'libs'
        }
    }
}

dependencies {
    compile(name:'tensorflow-lite', ext:'aar')
}

My current code and model requires implementations below to work correctly.

implementation 'org.tensorflow:tensorflow-lite:2.3.0'
implementation 'org.tensorflow:tensorflow-lite-gpu:2.3.0'
implementation 'org.tensorflow:tensorflow-lite-select-tf-ops:0.0.0-nightly'

The problem comes when the above three implementation or commenting two except implementation 'org.tensorflow:tensorflow-lite-gpu:2.3.0' gives errors,

cannot find symbol class Interpreter`
cannot find symbol class NnApiDelegate
error: package org.tensorflow.lite.nnapi does not exist

This error is not show when fat_apk is built using command, bazel build -c opt --fat_apk_cpu=arm64-v8a,armeabi-v7a --host_crosstool_top=@bazel_tools//tools/cpp:toolchain tensorflow/lite/java:tensorflow-lite. But using this method there is no tensorflow-lite-select-tf-ops.aar in /tensorflow_src/bazel-bin/tensorflow/lite/java/ folder.

Using the previous with above fat_apk tensorflow-lite aar it crashes with error,

java.lang.UnsatisfiedLinkError: dlopen failed: cannot locate symbol "_ZNK6google8protobuf7Message11GetTypeNameEv" referenced by "/data/app/com.example.myapp-PkLKD5MHuq9TB2R3mDgNcA==/base.apk!/lib/arm64-v8a/libtensorflowlite_flex_jni.so"...

My main goal is reduce size of select-ft-ops as I noticed it reduces app size by 60+mb. If there is an alternate method please suggest. Also looking for building instructions of tensorflow-lite-gpu aar.

Provide the exact sequence of commands / steps that you executed before running into the problem

• Setup docker in Ubuntu 18.04 and download dockerfile from here.
• Follow commands from Set up build environment using Docker section from here.
• Run ./configure.
• Use command, bash tensorflow/lite/tools/build_aar.sh --input_models=custom_model.tflite --target_archs=arm64-v8a,armeabi-v7a.
• Get the aar files from tmp folder as mentioned here in section Building the Android AAR and paste in libs folder of android project.
• Paste following,

allprojects {
    repositories {
        jcenter()
        flatDir {
            dirs 'libs'
        }
    }
}

dependencies {
    compile(name:'tensorflow-lite', ext:'aar')
    compile(name:'tensorflow-lite-select-tf-ops', ext:'aar')
}

• Commenting out following,

//implementation 'org.tensorflow:tensorflow-lite:2.3.0'
//implementation 'org.tensorflow:tensorflow-lite-select-tf-ops:0.0.0-nightly'

Any other info / logs Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.