TFLite:编译(app, so,jar, aar)

bazel 编译app

bazel build --cxxopt=--std=c++11 //tensorflow/contrib/lite/java/demo/app/src/main:TfLiteCameraDemo

这种方式编译的app是源码编译生成的AAR,jni还是下载下来的？

bazel 编译tensorflowlite库

bazel build --cxxopt='--std=c++11' //tensorflow/contrib/lite/java:tensorflowlite \
--crosstool_top=//external:android/crosstool \
--host_crosstool_top=@bazel_tools//tools/cpp:toolchain \
--cpu=armeabi

--cpu=armeabi
--cpu=armeabi-v7a
--cpu=arm64-v8a
--cpu=mips
--cpu=mips64
--cpu=x86
--cpu=x86_64

bazel-bin/tensorflow/contrib/lite/java/libtensorflowlite_jni.so
bazel-bin/tensorflow/contrib/lite/java/libtensorflowlitelib.jar

通过这种方式生成的文件，怎样编译到应用中去？

To build a standalone cc_binary or cc_library for Android without using an android_binary, use the --crosstool_top, --cpu and --host_crosstool_top flags.

For example:

bazel build //my/cc/jni:target \
      --crosstool_top=@androidndk//:default_crosstool \
      --cpu=<abi> \
      --host_crosstool_top=@bazel_tools//tools/cpp:toolchain

使用自定义 TensorFlow Lite 版本

bazel build --cxxopt='--std=c++11' -c opt        \
  --fat_apk_cpu=x86,x86_64,arm64-v8a,armeabi-v7a   \
  //tensorflow/contrib/lite/java:tensorflow-lite
   creating: lib/
   creating: lib/arm64-v8a/
  inflating: lib/arm64-v8a/libtensorflowlite_jni.so  
   creating: lib/armeabi-v7a/
  inflating: lib/armeabi-v7a/libtensorflowlite_jni.so  
   creating: lib/x86/
  inflating: lib/x86/libtensorflowlite_jni.so  
   creating: lib/x86_64/
  inflating: lib/x86_64/libtensorflowlite_jni.so  
  adding: jni/arm64-v8a/libtensorflowlite_jni.so (deflated 63%)
  adding: jni/armeabi-v7a/libtensorflowlite_jni.so (deflated 53%)
  adding: jni/x86/libtensorflowlite_jni.so (deflated 65%)
  adding: jni/x86_64/libtensorflowlite_jni.so (deflated 65%)
Target //tensorflow/contrib/lite/java:tensorflow-lite up-to-date:
  bazel-genfiles/tensorflow/contrib/lite/java/tensorflow-lite.aar

NNAPI的支持，并不是那么简单

// ASharedMemory_create was added in Android 8.0, so safe to use with NNAPI
// which was added in 8.1.
static void* handle = loadLibrary("libandroid.so");
libandroid.so是怎么生成的？

static void* handle = loadLibrary("libneuralnetworks.so");
编译文件并没有提供怎么生成libneuralnetworks.so
应该是和文件nnapi_delegate.cc有关

看懂tensorflowlite框架源码

要是你有tensorflow训练模型的知识，看懂tensorflowlite框架源码（主要就是interpreter.cc、model.cc）