Tensorflowlite移植ARM平台iMX6
一、LINUX环境下操作：
1.安装交叉编译SDK （仅针对该型号：i.MX6,不同芯片需要对应的交叉编译SDK）
编译方法参考：手动编译用于i.MX6系列的交叉编译SDK
2.下载Tensorflow
git clone https://github.com/tensorflow/tensorflow.git
cd tensorflow
git checkout r1.10
Tensorflow与Bazel编译器（及CUDA，CUDNN）之间需要对应，否则会有兼容性问题。
tensorflowr1.10 python 2.7,3.6 Bazel:0.18.0-0.19.2
tensorflowr1.12 python 2.7,3.6 Bazel:0.18.0-0.19.2
tensorflowr1.14 python 2.7,3.6 Bazel:0.24.0 - 0.25.2
3、下载并安装编译工具Bazel
安装依赖包：
sudo apt-get install pkg-config zip g++ zilb1g-dev unzip
下载Bazel包：
wget https://github.com/bazelbuild/bazel/releases/download/0.18.1/bazel-0.181-installer-linux-86_64.sh
安装Bazel：
chmod +x bazel-0.18.1-installer-linux-86_64.sh
./bazel-0.18.1-installer-linux-86_64.sh --user
设置环境变量：
sudo vi ~/.bashrc，在文件最后添加：export PATH=$PATH":~/bin"
source ~/.bashrc
（如果仅仅是测试DEMO在ARM板上使用，可直接跳过4，5，6，7，8步，直接进行第9步）
4、编译配置：
在Tensorflow源码根目录运行：
./configure （编译LINUX平台时使用默认设置：-march=native，编译ARM平台时需设置成相应值：-march=armv7-a）
5、编译pip:
bazel build --config=opt //tensorflow/tools/pip_package:build_pip_package
7、编译包：
./bazel-bin/tensorflow/tools/pip__package/build_pip_package /tmp/tensorflow_pkg
8、安装包：
pip install /tmp/tensorflow_pkg/tensorflow-version-tags.whl
9、下载依赖库：
./tensorflow/contrib/lite/tools/make/download_dependencies.sh(不同版本，位置略有不同，本文路径为r1.10版本)
10、编译Tensorflow Lite:
方法1：（参考网上的，但我调试了几天，并未生成该库，可行性待验证！）
目前Tensorflow仅支持树莓派： ./tensorflow/contrib/lite/tools/make/build_rpi_lib.sh，该脚本的目标编译平台是ARMv7，即使目标平台是ARMv8，也不要更改。因为设置编译平台为ARMv7可以优化编译，提高运行速度。
所以，需要针对iMX6，制作一份build_imx6_lib.sh
1 set -e 2 3 SCRIPT_DIR="$(cd "$(dirname "$(BASH_SOURCE[0]}")" && pwd)" 4 cd "$SCRIPT_DIR/../../.." 5 6 CC_PREFIX=arm-poky-linux-gnueabi- make -j 3 -f tensorflow/contrib/lite/Makefile TARGET-imx6 TARGET_ARCH=armv7-a
生成静态库位置为：
./tensorflow/contrib/lite/toos/make/gen/rpi_arm7l/lib/libtensorflow-lite.a静态库。
方法2：（使用ARM（iMX6）交叉编译包，编译出来的包可以用于iMX6平台）
1.打开Tensorflow/contrib/lite/kernels/internal/BUILD
1 config_setting( 2 name = "armv7a", 3 values = {"cpu":"armv7a",}, 4 ) 5 6 "armv7a":[ 7 "-O3", 8 "-mfpu=vfpv3", 9 "-mfloat-abi=hard", 10 ], 11 12 add code above to NEON_FLAGS_IF_APPLICABLE 13 14 15 ":armv6":[ 16 17 ":neon_tensor_utils", 18 ], 19 ":armv7a":[ 20 ":neon_tensor_utils", 21 ], 22 23 add code above to cc_library tensor_utils select options
2.在根目录下创建一个文件：build_armv7_tflite.sh
1 #!/bin/bash 2 3 bazel build --copt="-fPIC" --copt="-march=armv6" \ 4 --copt="-mfloat-abi=hard" \ 5 --copt="-mfpu=vfpv3" \ 6 --copt="-funsafe-math-optimizations" \ 7 --copt="-Wno-unused-function" \ 8 --copt="-Wno-sign-compare" \ 9 --copt="-ftree-vectorize" \ 10 --copt="-fomit-frame-pointer" \ 11 --cxxopt='--std=c++11' \ 12 --cxxopt="-Wno-maybe-uninitialized" \ 13 --cxxopt="-Wno-narrowing" \ 14 --cxxopt="-Wno-unused" \ 15 --cxxopt="-Wno-comment" \ 16 --cxxopt="-Wno-unused-function" \ 17 --cxxopt="-Wno-sign-compare" \ 18 --cxxopt="-funsafe-math-optimizations" \ 19 --linkopt="-lstdc++" \ 20 --linkopt="-mfloat-abi=hard" \ 21 --linkopt="-mfpu=vfpv3" \ 22 --linkopt="-funsafe-math-optimizations" \ 23 --verbose_failures \ 24 --strip=always \ 25 --crosstool_top=//armv6-compiler:toolchain --cpu=armv7 --config=opt \ 26 33 tensorflow/contrib/lite/examples/label_image
3.编译该文件build_armv7_tflite.sh，会碰到一个错误：.../.../read-ld:unrecognized options : --icf=all
解决方法：找到文件./build_def.bzl ,打开，去除所有--icf=all标识的信息即可。
编译后，会在bazel_bin下生成指定的文件label_image。
4.生成头和库文件：
libbuiltin_ops.a libframework.a libneon_tensor_utils.a libquantization_util.a libtensor_utils.a
libcontext.a libfarmhash.a libgemm_support.a libportable_tensor_utils.a libstring_util.a
11、编译模型：
默认情况下label_image并未编译进去，需要修改Makefile，可参考minimal APK,主要修改以下三部分内容：
LABELIMAGE_SRCS
LABELIMAGE_BINARY
LABEL_OBJS
1 # Find where we're running from, so we can store generated files here. 2 ifeq ($(origin MAKEFILE_DIR), undefined) 3 MAKEFILE_DIR := $(shell dirname $(realpath $(lastword $(MAKEFILE_LIST)))) 4 endif 5 6 # Try to figure out the host system 7 HOST_OS := 8 ifeq ($(OS),Windows_NT) 9 HOST_OS = WINDOWS 10 else 11 UNAME_S := $(shell uname -s) 12 ifeq ($(UNAME_S),Linux) 13 HOST_OS := LINUX 14 endif 15 ifeq ($(UNAME_S),Darwin) 16 HOST_OS := OSX 17 endif 18 endif 19 20 #HOST_ARCH := $(shell if [[ $(shell uname -m) =~ i[345678]86 ]]; then echo x86_32; else echo $(shell uname -m); fi) 21 22 # Self-hosting 23 TARGET_ARCH := ${HOST_ARCH} 24 CROSS := imx6 25 $(warning "CROSS :$(CROSS) HOST_ARCH:$(HOST_ARCH) TARGET_ARCH:$(TARGET_ARCH) TARGET_TOOLCHAIN_PREFIX:$(TARGET_TOOLCHAIN_PREFIX)") 26 # Cross compiling 27 ifeq ($(CROSS),imx6) 28 TARGET_ARCH := armv7-a 29 TARGET_TOOLCHAIN_PREFIX := arm-poky-linux-gnueabi- 30 endif 31 32 ifeq ($(CROSS),rpi) 33 TARGET_ARCH := armv7l 34 TARGET_TOOLCHAIN_PREFIX := arm-linux-gnueabihf- 35 endif 36 37 ifeq ($(CROSS),riscv) 38 TARGET_ARCH := riscv 39 TARGET_TOOLCHAIN_PREFIX := riscv32-unknown-elf- 40 endif 41 ifeq ($(CROSS),stm32f7) 42 TARGET_ARCH := armf7 43 TARGET_TOOLCHAIN_PREFIX := arm-none-eabi- 44 endif 45 ifeq ($(CROSS),stm32f1) 46 TARGET_ARCH := armm1 47 TARGET_TOOLCHAIN_PREFIX := arm-none-eabi- 48 endif 49 50 # Where compiled objects are stored. 51 OBJDIR := $(MAKEFILE_DIR)/gen/obj/ 52 BINDIR := $(MAKEFILE_DIR)/gen/bin/ 53 LIBDIR := $(MAKEFILE_DIR)/gen/lib/ 54 GENDIR := $(MAKEFILE_DIR)/gen/obj/ 55 56 LIBS := 57 ifeq ($(TARGET_ARCH),x86_64) 58 CXXFLAGS += -fPIC -DGEMMLOWP_ALLOW_SLOW_SCALAR_FALLBACK -pthread # -msse4.2 59 endif 60 61 62 ifeq ($(TARGET_ARCH),armv7-a) 63 CXXFLAGS += -pthread -fPIC 64 LIBS += -ldl 65 endif 66 67 ifeq ($(TARGET_ARCH),armv7l) 68 CXXFLAGS += -mfpu=neon -pthread -fPIC 69 LIBS += -ldl 70 endif 71 72 ifeq ($(TARGET_ARCH),riscv) 73 # CXXFLAGS += -march=gap8 74 CXXFLAGS += -DTFLITE_MCU 75 LIBS += -ldl 76 BUILD_TYPE := micro 77 endif 78 79 ifeq ($(TARGET_ARCH),armf7) 80 CXXFLAGS += -DGEMMLOWP_ALLOW_SLOW_SCALAR_FALLBACK -DTFLITE_MCU 81 CXXFLAGS += -fno-rtti -fmessage-length=0 -fno-exceptions -fno-builtin -ffunction-sections -fdata-sections 82 CXXFLAGS += -funsigned-char -MMD 83 CXXFLAGS += -mcpu=cortex-m7 -mthumb -mfpu=fpv5-sp-d16 -mfloat-abi=softfp 84 CXXFLAGS += '-std=gnu++11' '-fno-rtti' '-Wvla' '-c' '-Wall' '-Wextra' '-Wno-unused-parameter' '-Wno-missing-field-initializers' '-fmessage-length=0' '-fno-exceptions' '-fno-builtin' '-ffunction-sections' '-fdata-sections' '-funsigned-char' '-MMD' '-fno-delete-null-pointer-checks' '-fomit-frame-pointer' '-Os' 85 LIBS += -ldl 86 BUILD_TYPE := micro 87 endif 88 ifeq ($(TARGET_ARCH),armm1) 89 CXXFLAGS += -DGEMMLOWP_ALLOW_SLOW_SCALAR_FALLBACK -mcpu=cortex-m1 -mthumb -DTFLITE_MCU 90 CXXFLAGS += -fno-rtti -fmessage-length=0 -fno-exceptions -fno-builtin -ffunction-sections -fdata-sections 91 CXXFLAGS += -funsigned-char -MMD 92 LIBS += -ldl 93 endif 94 95 # Settings for the host compiler. 96 #CXX := $(CC_PREFIX) ${TARGET_TOOLCHAIN_PREFIX}g++ 97 #CXX := ${TARGET_TOOLCHAIN_PREFIX}g++ -march=armv7-a -mfpu=neon -mfloat-abi=hard -mcpu=cortex-a9 --sysroot=/opt/fsl-imx-x11/4.1.15-2.1.0/sysroots/x86_64-pokysdk-linux/usr/bin/arm-poky-linux-gnueabi 98 CXXFLAGS += -O3 -DNDEBUG 99 CCFLAGS := ${CXXFLAGS} 100 #CXXFLAGS += --std=c++11 101 CXXFLAGS += --std=c++0x #wly 102 #CC := ${TARGET_TOOLCHAIN_PREFIX}gcc 103 #AR := ${TARGET_TOOLCHAIN_PREFIX}ar 104 CFLAGS := 105 LDOPTS := 106 LDOPTS += -L/usr/local/lib 107 ARFLAGS := -r 108 109 INCLUDES := \ 110 -I. \ 111 -I$(MAKEFILE_DIR)/../../../ \ 112 -I$(MAKEFILE_DIR)/downloads/ \ 113 -I$(MAKEFILE_DIR)/downloads/eigen \ 114 -I$(MAKEFILE_DIR)/downloads/gemmlowp \ 115 -I$(MAKEFILE_DIR)/downloads/neon_2_sse \ 116 -I$(MAKEFILE_DIR)/downloads/farmhash/src \ 117 -I$(MAKEFILE_DIR)/downloads/flatbuffers/include \ 118 -I$(GENDIR) 119 # This is at the end so any globally-installed frameworks like protobuf don't 120 # override local versions in the source tree. 121 #INCLUDES += -I/usr/local/include 122 INCLUDES += -I/usr/include 123 LIBS += \ 124 -lstdc++ \ 125 -lpthread \ 126 -lm \ 127 -lz 128 129 # If we're on Linux, also link in the dl library. 130 ifeq ($(HOST_OS),LINUX) 131 LIBS += -ldl 132 endif 133 134 include $(MAKEFILE_DIR)/ios_makefile.inc 135 ifeq ($(CROSS),rpi) 136 #include $(MAKEFILE_DIR)/rpi_makefile.inc 137 endif 138 ifeq ($(CROSS),imx6) 139 include $(MAKEFILE_DIR)/imx6_makefile.inc 140 endif 141 # This library is the main target for this makefile. It will contain a minimal 142 # runtime that can be linked in to other programs. 143 LIB_NAME := libtensorflow-lite.a 144 LIB_PATH := $(LIBDIR)$(LIB_NAME) 145 146 # A small example program that shows how to link against the library. 147 MINIMAL_PATH := $(BINDIR)minimal 148 LABEL_IMAGE_PATH :=$(BINDIR)label_image 149 150 MINIMAL_SRCS := \ 151 tensorflow/contrib/lite/examples/minimal/minimal.cc 152 MINIMAL_OBJS := $(addprefix $(OBJDIR), \ 153 $(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(MINIMAL_SRCS)))) 154 155 156 LABEL_IMAGE_SRCS := \ 157 tensorflow/contrib/lite/examples/label_image/label_image.cc \ 158 tensorflow/contrib/lite/examples/label_image/bitmap_helpers.cc 159 LABEL_IMAGE_OBJS := $(addprefix $(OBJDIR), \ 160 $(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(LABEL_IMAGE_SRCS)))) 161 162 # What sources we want to compile, must be kept in sync with the main Bazel 163 # build files. 164 165 PROFILER_SRCS := \ 166 tensorflow/contrib/lite/profiling/time.cc 167 PROFILE_SUMMARIZER_SRCS := \ 168 tensorflow/contrib/lite/profiling/profile_summarizer.cc \ 169 tensorflow/core/util/stats_calculator.cc 170 171 CORE_CC_ALL_SRCS := \ 172 $(wildcard tensorflow/contrib/lite/*.cc) \ 173 $(wildcard tensorflow/contrib/lite/*.c) 174 ifneq ($(BUILD_TYPE),micro) 175 CORE_CC_ALL_SRCS += \ 176 $(wildcard tensorflow/contrib/lite/kernels/*.cc) \ 177 $(wildcard tensorflow/contrib/lite/kernels/internal/*.cc) \ 178 $(wildcard tensorflow/contrib/lite/kernels/internal/optimized/*.cc) \ 179 $(wildcard tensorflow/contrib/lite/kernels/internal/reference/*.cc) \ 180 $(PROFILER_SRCS) \ 181 $(wildcard tensorflow/contrib/lite/kernels/*.c) \ 182 $(wildcard tensorflow/contrib/lite/kernels/internal/*.c) \ 183 $(wildcard tensorflow/contrib/lite/kernels/internal/optimized/*.c) \ 184 $(wildcard tensorflow/contrib/lite/kernels/internal/reference/*.c) \ 185 $(wildcard tensorflow/contrib/lite/downloads/farmhash/src/farmhash.cc) \ 186 $(wildcard tensorflow/contrib/lite/downloads/fft2d/fftsg.c) 187 endif 188 # Remove any duplicates. 189 CORE_CC_ALL_SRCS := $(sort $(CORE_CC_ALL_SRCS)) 190 CORE_CC_EXCLUDE_SRCS := \ 191 $(wildcard tensorflow/contrib/lite/*test.cc) \ 192 $(wildcard tensorflow/contrib/lite/*/*test.cc) \ 193 $(wildcard tensorflow/contrib/lite/*/*/*test.cc) \ 194 $(wildcard tensorflow/contrib/lite/*/*/*/*test.cc) \ 195 $(wildcard tensorflow/contrib/lite/kernels/test_util.cc) \ 196 $(MINIMAL_SRCS) \ 197 $(LABEL_IMAGE_SRCS) 198 ifeq ($(BUILD_TYPE),micro) 199 CORE_CC_EXCLUDE_SRCS += \ 200 tensorflow/contrib/lite/model.cc \ 201 tensorflow/contrib/lite/nnapi_delegate.cc 202 endif 203 # Filter out all the excluded files. 204 TF_LITE_CC_SRCS := $(filter-out $(CORE_CC_EXCLUDE_SRCS), $(CORE_CC_ALL_SRCS)) 205 # File names of the intermediate files target compilation generates. 206 TF_LITE_CC_OBJS := $(addprefix $(OBJDIR), \ 207 $(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(TF_LITE_CC_SRCS)))) 208 LIB_OBJS := $(TF_LITE_CC_OBJS) 209 210 # Benchmark sources 211 BENCHMARK_SRCS_DIR := tensorflow/contrib/lite/tools/benchmark 212 BENCHMARK_ALL_SRCS := $(TFLITE_CC_SRCS) \ 213 $(wildcard $(BENCHMARK_SRCS_DIR)/*.cc) \ 214 $(PROFILE_SUMMARIZER_SRCS) 215 216 BENCHMARK_SRCS := $(filter-out \ 217 $(wildcard $(BENCHMARK_SRCS_DIR)/*_test.cc), \ 218 $(BENCHMARK_ALL_SRCS)) 219 220 BENCHMARK_OBJS := $(addprefix $(OBJDIR), \ 221 $(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(BENCHMARK_SRCS)))) 222 223 # For normal manually-created TensorFlow C++ source files. 224 $(OBJDIR)%.o: %.cc 225 @mkdir -p $(dir $@) 226 $(CXX) $(CXXFLAGS) $(INCLUDES) -c $< -o $@ 227 # For normal manually-created TensorFlow C++ source files. 228 $(OBJDIR)%.o: %.c 229 @mkdir -p $(dir $@) 230 $(CC) $(CCFLAGS) $(INCLUDES) -c $< -o $@ 231 232 # The target that's compiled if there's no command-line arguments. 233 all: $(LIB_PATH) $(MINIMAL_PATH) $(LABEL_IMAGE_PATH) $(BENCHMARK_BINARY) 234 235 # The target that's compiled for micro-controllers 236 micro: $(LIB_PATH) 237 238 # Gathers together all the objects we've compiled into a single '.a' archive. 239 $(LIB_PATH): $(LIB_OBJS) 240 @mkdir -p $(dir $@) 241 $(AR) $(ARFLAGS) $(LIB_PATH) $(LIB_OBJS) 242 243 $(MINIMAL_PATH): $(MINIMAL_OBJS) $(LIB_PATH) 244 @mkdir -p $(dir $@) 245 $(CXX) $(CXXFLAGS) $(INCLUDES) \ 246 -o $(MINIMAL_PATH) $(MINIMAL_OBJS) \ 247 $(LIBFLAGS) $(LIB_PATH) $(LDFLAGS) $(LIBS) 248 249 # $(LABEL_IMAGE_PATH): $(LABEL_IMAGE_OBJS) $(LIBS) 250 251 $(LABEL_IMAGE_PATH) : $(LABEL_IMAGE_OBJS) $(LIB_PATH) 252 @mkdir -p $(dir $@) 253 $(CXX) $(CXXFLAGS) $(INCLUDES) \ 254 -o $(LABEL_IMAGE_PATH) $(LABEL_IMAGE_OBJS)\ 255 $(LIBFLAGS) $(LIB_PATH) $(LDFLAGS) $(LIBS) 256 257 258 $(BENCHMARK_LIB) : $(LIB_PATH) $(BENCHMARK_OBJS) 259 @mkdir -p $(dir $@) 260 $(AR) $(ARFLAGS) $(BENCHMARK_LIB) $(LIB_OBJS) $(BENCHMARK_OBJS) 261 262 benchmark_lib: $(BENCHMARK_LIB) 263 $(info $(BENCHMARK_BINARY)) 264 $(BENCHMARK_BINARY) : $(BENCHMARK_LIB) 265 @mkdir -p $(dir $@) 266 $(CXX) $(CXXFLAGS) $(INCLUDES) \ 267 -o $(BENCHMARK_BINARY) \ 268 $(LIBFLAGS) $(BENCHMARK_LIB) $(LDFLAGS) $(LIBS) 269 270 benchmark: $(BENCHMARK_BINARY) 271 272 # Gets rid of all generated files. 273 clean: 274 rm -rf $(MAKEFILE_DIR)/gen 275 276 # Gets rid of target files only, leaving the host alone. Also leaves the lib 277 # directory untouched deliberately, so we can persist multiple architectures 278 # across builds for iOS and Android. 279 cleantarget: 280 rm -rf $(OBJDIR) 281 rm -rf $(BINDIR) 282 283 $(DEPDIR)/%.d: ; 284 .PRECIOUS: $(DEPDIR)/%.d 285 286 -include $(patsubst %,$(DEPDIR)/%.d,$(basename $(TF_CC_SRCS)))
执行： ./tensorflow/contrib/lite/tools/make/build_imx6_lib.sh
编译过程中可能会出现一些问题，依次解决即可！
编译完成后，在./tensorflow/contrib/lite/tools/make/gen/rpi_arm7l/bin 目录下生成可执行文件label_img
附上自己碰到的一些问题：
1、a/tensorflow/contrib/lite/tools/benchmark/benchmark_params.h
1 diff --git a/tensorflow/contrib/lite/profiling/profile_summarizer.cc b/tensorflow/contrib/lite/profiling/profile_summarizer.cc 2 old mode 100644 3 new mode 100755 4 index c37a096..590cd21 5 --- a/tensorflow/contrib/lite/profiling/profile_summarizer.cc 6 +++ b/tensorflow/contrib/lite/profiling/profile_summarizer.cc 7 @@ -83,7 +83,8 @@ OperatorDetails GetOperatorDetails(const tflite::Interpreter& interpreter, 8 OperatorDetails details; 9 details.name = op_name; 10 if (profiling_string) { 11 - details.name += ":" + string(profiling_string); 12 + //wly 13 + details.name += ":" + std::string(profiling_string); 14 } 15 details.inputs = GetTensorNames(interpreter, inputs); 16 details.outputs = GetTensorNames(interpreter, outputs);
2、./tensorflow/contrib/lite/profiling/profile_summarizer.cc
1 diff --git a/tensorflow/contrib/lite/tools/benchmark/benchmark_params.h b/tensorflow/contrib/lite/tools/benchmark/benchmark_params.h 2 old mode 100644 3 new mode 100755 4 index 33448dd..e7f63ff 5 --- a/tensorflow/contrib/lite/tools/benchmark/benchmark_params.h 6 +++ b/tensorflow/contrib/lite/tools/benchmark/benchmark_params.h 7 @@ -47,7 +47,8 @@ class BenchmarkParam { 8 virtual ~BenchmarkParam() {} 9 BenchmarkParam(ParamType type) : type_(type) {} 10 11 - private: 12 + //private: 13 + public: //wly 14 static void AssertHasSameType(ParamType a, ParamType b); 15 template <typename T> 16 static ParamType GetValueType();
3、./tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h
1 diff --git a/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h b/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h 2 old mode 100644 3 new mode 100755 4 index 2d40f17..43c54dc 5 --- a/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h 6 +++ b/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h 7 @@ -24,8 +24,9 @@ limitations under the License. 8 #include <type_traits> 9 10 - #include "third_party/eigen3/Eigen/Core" 11 +//#include "third_party/eigen3/Eigen/Core"" 12 #include "tensorflow/contrib/lite/kernels/internal/common.h" 13 #include "tensorflow/contrib/lite/kernels/internal/quantization_util.h" 14 #include "tensorflow/contrib/lite/kernels/internal/round.h"
4、需要下载包：protobuf
5、将c++11 改为c++0x
12、在PC上测试label_image
./label_image -v 1 -m ./mobilenet_v1_1.0_224.tflite -i ./grace_hopper.jpg -l ./imagenet_slim_labels.txt
报错信息： bash ./label_image: cannot execute binary file:Exec format error
原因有两个：
一是GCC编译时多加了一个-C，生成了二进制文件；
解决方法:找到GCC编译处，去除-C选项。
二就是编译环境不同（平台芯片不一致）导致
解决方法：需要在对应平台编译。
在tensorflow根目录执行：bazel build tensorflow/examples/label_image:label_image
如果是第一次编译，时间较久；
编译完成后，生成可执行文件：bazel_bin/tensorflow/examples/label_image/label_image
本地测试该文件：
拷贝label_image和libtensorflow_framework.so到tensorflow/examples/label_image下
(第一次测试时，未拷贝libtensorflow_framework.so，直接提示：error while loading shared libraryies:libtensorflow_framework.so:cannot open shared object file:No such file or directory)
再次运行./tensorflow/examples/label_image/label_image
显示结果：military uniform(653):0.834306
mortarboard(668):0.0218695
academic gown(401):0.0103581
pickelhaube(716):0.00800814
bulletproof vest(466):0.00535084
说明测试OK！
以下操作在ARM板子上：
1、拷贝生成的label_image到板子上（bazel_bin/tensorflow/contrib/lite/examples/label_image/）(1.7M)
拷贝生成的label_image到板子上（bazel_bin/tensorflow/examples/label_image）(52.2M)
2、拷贝图片./tensorflow/examples/label_image/data/grace_hopper.jpg到板子上
3、下载模型mobilenet_quant_v1_224.tflite
（如果下载其它模型，可参考该文件：/tensorflow/contrib/lite/g3doc/models.md）
4、下载模型所需文件：
curl -L "https://storage.googleapis.com/download.tensorflow.org/models/inception_v3_2016_08_28_frozen.pb.tar.gz" | tar -C tensorflow/examples/label_image/data -xz
5、拷贝libtensorflow_framework.so到板子上，文件位于目录bazel_bin/tensorflow/下（14M）
4、运行label_image
确保如下所需文件都已完成：
（拷贝文件1：label_image）
（拷贝文件2：grace_hopper.jpg）
（拷贝文件3：mobilenet_quant_v1_224.tflite）
（拷贝文件4：imagenet_slim_labels.txt）
（拷贝文件5：libtensorflow_framework.so）
ARM板上常用操作：
mount /dev/sba1 /mnt/usb
ls /mnt/usb
cp /mnt/usb/ .
执行脚本：
./label_image
如果出现-sh: ./label_image: not found，可能是编译器不一致导致。
尝试方法1：重定向:ln -s ld-linux.so.3 ld-linux-armhf.so.3
新报错：./label_image:/lib/libm.so.6: version 'GLIBC_2.27' not found (required by ./label_image)
出现以下信息，说明已经OK，恭喜你!
