tensorflow-lite-select-tf-ops 和从源 tflite 模型构建的 tensorflow-lite AAR 找不到 org.tensorflow.lite.Interpreter

Question

我按照 tensorflow android 指南从源代码生成 tensorflow-lite.aar 和 tensorflow-lite-select-tf-ops.aar。

https://www.tensorflow.org/lite/guide/build_android

我当前的应用程序适用于，

dependencies {
    ...
    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:2.3.0'
}

我希望减小 tensorflow-lite-select-tf-ops.aar 的大小以减小应用程序 apk 的大小。

我正在按照 build_aar.sh 流程来减小我的 tflite 模型的大小。使用的命令，

bash tensorflow/lite/tools/build_aar.sh --input_models=custom_model.tflite --target_archs=arm64-v8a

这会在 custom_model.tflite 的 bazel-bin/tmp 文件夹中生成 tensorflow-lite.aar 和 tensorflow-lite-select-tf-ops.aar。 tflite 模型已 FP16 量化。

使用新生成的 aar 文件实现，

dependencies {
    ...
//    implementation 'org.tensorflow:tensorflow-lite-select-tf-ops:2.3.0'
//    implementation 'org.tensorflow:tensorflow-lite:2.3.0'
    implementation 'org.tensorflow:tensorflow-lite-gpu:2.3.0'

    implementation files('libs/tensorflow-lite.aar')
//    implementation files('libs/tensorflow-lite-gpu.aar')
    implementation files('libs/tensorflow-lite-select-tf-ops.aar')
}

但是当它们包含在 android 应用程序编译错误时发生。

C:\folder\myapp\app\src\main\java\com\example\mapp\tflite\YoloV4Classifier.java:19: error: cannot find symbol
import org.tensorflow.lite.Interpreter;
                          ^
  symbol:   class Interpreter
  location: package org.tensorflow.lite

使用 bazel build -c opt --fat_apk_cpu=arm64-v8a --host_crosstool_top=@bazel_tools//tools/cpp:toolchain tensorflow/lite/java:tensorflow-lite 构建 tensorflow-lite.aar 时，此问题消失。但这引入了 UnsatisfiedLinkError: dlopen failed: cannot locate symbol with previous tensorflow-lite-select-tf-ops.aar and new built tensorflow-lite.aar with above command.

如果我把之前的 tensorflow-lite.aar 生成自 custom_model.tflite 和新的 tensorflow-lite.aar 生成的 bazel build -c opt --fat_apk_cpu=arm64-v8a --host_crosstool_top=@bazel_tools//tools/cpp:toolchain tensorflow/lite/java:tensorflow-lite 那么就会出现构建错误，

More than one file was found with OS independent path 'lib/arm64-v8a/libtensorflowlite_jni.so'

我按照基于非输入 tflite 模型的命令从源代码生成 tensorflow-lite.aar、tensorflow-lite-gpu.aar、tensorflow-lite-select-tf-ops.aar，但应用程序大小变得非常大。我试过 other ways 但每次都会出现一种新的错误。

Answer 1

问题已解决，如下所述，

https://github.com/tensorflow/tensorflow/issues/45153

https://github.com/tensorflow/tensorflow/issues/46989

https://github.com/tensorflow/tensorflow/issues/45488

tensorflow-lite-select-tf-ops 和从源 tflite 模型构建的 tensorflow-lite AAR 找不到 org.tensorflow.lite.Interpreter

tensorflow-lite-select-tf-ops and tensorflow-lite AAR built from source tflite model cannot find org.tensorflow.lite.Interpreter

android

build.gradle

aar

tensorflow

tensorflow-lite