Environment Setup

Cross-compilation SDK Installation

Install the cross-compilation tooclhain specific to the i.MX6 platform. Different chips require corresponding toolchains.

TensorFlow Source Retrieval

Clone the TensorFlow repository:

git clone https://github.com/tensorflow/tensorflow.git
cd tensorflow
git checkout r1.10

Ensure compatibility between TensorFlow version and Bazel compiler. For r1.10, use Python 2.7 or 3.6 with Bazel 0.18.0-0.19.2.

Bazel Compiler Installation

Install required dependencies:

sudo apt-get install pkg-config zip g++ zlib1g-dev unzip

Download and install Bazel:

wget https://github.com/bazelbuild/bazel/releases/download/0.18.1/bazel-0.18.1-installer-linux-x86_64.sh
chmod +x bazel-0.18.1-installer-linux-x86_64.sh
./bazel-0.18.1-installer-linux-x86_64.sh --user

Add to environment variables:

export PATH=$PATH:~/bin

Configuration and Build Process

Configure Build Settings

In the TensorFlow root directory:

./configure

Set appropriate flags for ARM architecture when targeting i.MX6:

--config=opt --copt=-march=armv7-a --copt=-mfpu=neon --copt=-mfloat-abi=hard

Compile Python Package

bazel build --config=opt //tensorflow/tools/pip_package:build_pip_package

Generate wheel package:

./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

Install package:

pip install /tmp/tensorflow_pkg/tensorflow-*.whl

Download Dependencies

./tensorflow/contrib/lite/tools/make/download_dependencies.sh

ARM Platform Compilation

Create Custom Build Script

Create a build script for i.MX6 platform:

#!/bin/bash
set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
cd "$SCRIPT_DIR/../../.."

CC_PREFIX=arm-poky-linux-gnueabi- make -j 3 -f tensorflow/contrib/lite/Makefile TARGET=imx6 TARGET_ARCH=armv7-a

The resulting static library will be located at:

tensorflow/contrib/lite/tools/make/gen/rpi_arm7l/lib/libtensorflow-lite.a

Alternative Build Method Using Bazel

Modify BUILD Configuration

Edit tensorflow/contrib/lite/kernels/internal/BUILD to include ARMv7a optimizations:

config_setting(
    name = "armv7a",
    values = {"cpu": "armv7a"},
)

"armv7a": [
    "-O3",
    "-mfpu=vfpv3",
    "-mfloat-abi=hard",
],

Create Compilation Script

Create build_armv7_tflite.sh:

#!/bin/bash

bazel build --copt="-fPIC" \
    --copt="-march=armv6" \
    --copt="-mfloat-abi=hard" \
    --copt="-mfpu=vfpv3" \
    --copt="-funsafe-math-optimizations" \
    --copt="-Wno-unused-function" \
    --copt="-Wno-sign-compare" \
    --copt="-ftree-vectorize" \
    --copt="-fomit-frame-pointer" \
    --cxxopt='--std=c++11' \
    --cxxopt="-Wno-maybe-uninitialized" \
    --cxxopt="-Wno-narrowing" \
    --cxxopt="-Wno-unused" \
    --cxxopt="-Wno-comment" \
    --cxxopt="-Wno-unused-function" \
    --cxxopt="-Wno-sign-compare" \
    --cxxopt="-funsafe-math-optimizations" \
    --linkopt="-lstdc++" \
    --linkopt="-mfloat-abi=hard" \
    --linkopt="-mfpu=vfpv3" \
    --linkopt="-funsafe-math-optimizations" \
    --verbose_failures \
    --strip=always \
    --crosstool_top=//armv6-compiler:toolchain \
    --cpu=armv7 \
    --config=opt \
    tensorflow/contrib/lite/examples/label_image

Fix linker issues by removing --icf=all from build_def.bzl.

Model Compilation

Modify Makefile to include label_image binary:

LABEL_IMAGE_SRCS := \
tensorflow/contrib/lite/examples/label_image/label_image.cc \
tensorflow/contrib/lite/examples/label_image/bitmap_helpers.cc

LABEL_IMAGE_OBJS := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(LABEL_IMAGE_SRCS))))

all: $(LIB_PATH) $(MINIMAL_PATH) $(LABEL_IMAGE_PATH) $(BENCHMARK_BINARY)

Execute build script:

./tensorflow/contrib/lite/tools/make/build_imx6_lib.sh

Testing on Host System

Compile for host platform:

bazel build tensorflow/examples/label_image:label_image

Copy necessary files:

# Copy executable and shared library
cp bazel-bin/tensorflow/examples/label_image/label_image .
cp bazel-bin/tensorflow/libtensorflow_framework.so .

Run test:

./label_image -v 1 -m ./mobilenet_v1_1.0_224.tflite -i ./grace_hopper.jpg -l ./imagenet_slim_labels.txt

Deployment on Target Device

Transfer files to ARM board:

• label_image (1.7MB)
• grace_hopper.jpg
• mobilenet_quant_v1_224.tflite
• imagenet_slim_labels.txt
• libtensorflow_framework.so (14MB)

Mount and copy files:

mount /dev/sda1 /mnt/usb
ls /mnt/usb
cp /mnt/usb/* .

Execute application:

./label_image

If execution fails with "not found", ensure proper architecture compatibility and correct dynamic linker setup.