Table of contents

Preparation

Please refer here for the environment construction procedure.

Conversion

Convert the model.

Please change the parameter "setting.conf" of the argument according to the model to be converted.

$ cd /home/cvtool/conversion/tensorflow
$ ./tf_conversion.sh setting.conf

The model after conversion is output to the following directory.

${OUTPUT_DIR}/${NET_NAME}/${PARSER_OPTION}/[モデル名]

For models that use SSD, another binary is output to the following directory.

Models with SSD use two binary files in the application.

${OUTPUT_DIR}/${NET_NAME}/

Setting.conf

# Network Name
NET_NAME=inception_v4


# SSD Model or Not (0:not SSD, 1:SSD)
IS_SSD=0

# Path to Directory for (frozen) Protobuf File
PB_DIR=./sample/model

# Path to Directory for DRA Images
DRA_IMAGE_DIR=../../dra_img

# Path to Directory for Output Data
OUTPUT_DIR=./out

# Quantization Mode
#  FIX8  : Fixed-point  8bit
#  FIX16 : Fixed-point 16bit
#  MIX   : FIX8/FIX16 mixed
PARSER_OPTION=MIX

# Input Data Format (0:NHWC, 1:NCHW)
IN_DATA_FORMAT=0

# Input Data Channel
IN_DATA_CHANNEL=3

# Input Data Width
IN_DATA_WIDTH=299

# Input Data Height
IN_DATA_HEIGHT=299

# Input Data Mean Vector
IN_MEAN=127.5,127.5,127.5

# Input Data Scale
# IN_SCALE=1/Scale
IN_SCALE=127.5

# RGB or BGR (0:RGB, 1:BGR)
IS_BGR=1

# Input Nodes Name
IN_NODE=input

# Output Nodes Name
OUT_NODE=InceptionV4/Logits/Predictions

# (Need to Set When IS_SSD=1) Node Name as "priorbox"
PRIORBOX_NODE=Concatenate/concat

#cavalry version
#if not specified -> ""
CAVALRY_VER="2.1.7"

# Unique preprocess
# if use im2bin -> NONE
# if use unique preprocess -> script path
PREPRO=NONE
PREPRO_ARG=""

# Parser Data Format (0:NHWC, 1:NCHW)
PARSER_IN_DATA_FORMAT=1

# Input file data format
IN_DATA_FILEFORMAT=0,0,0,0

# Transpose indices(NONE:without transpose , 0,3,1,2:transpose (EX))
IN_DATA_TRANSPOSE=NONE

• NET_NAME: The name of network

  • Any name can be set.

• IS_SSD: Set 1, if target model uses SSD

• PB_DIR: Path to directory which includes frozen .pb files

  • All .pb files under the directory are converted.

• IMAGE_DIR: Path to directory that includes image files for optimizing quantization

  • Please put the directory image files for training. Recommended number of image files is 100 to 200.

  • Available image file format is what OpenCV can handle, for example, JPEG, PNG and so on.

  • Any resolution is available.

• OUTPUT_DIR: Path to directory to which the converted data is placed

• PARSER_OPTION: Quantization mode

  • Select from FIX8/FIX16/MIX (FIX8/FIX16 mixed).

• IN_DATA_FORMAT: Format of input data for target model (NHWC or NCHW)

• IN_DATA_CHANNEL: Number of input image channel for target model

• N_DATA_WIDTH: Width of input image for target model

• IN_DATA_HEIGHT: Height of input image for target model

• IN_MEAN: Normalization parameter (mean) of input image

  • Please refrain from using space between “,” as shown below if using numerical value.
    IN_MEAN=127.5,127.5,127.5

• IN_SCALE: Normalization parameter (scale) of input image

  • Please refrain from using space between “,” and only use “,” to separate values when setting different values for each channel.

• IS_BGR: Format of input image (RGB or BGR)

• IN_NODE: The name of input node for target network

  • When the following symbols are contained in the name of input node, conversion may not be successful.

    • | ; , ‘

• OUT_NODE: The name of output node for target network

  • If two or more nodes exists, separate layers by “,”.

  • When the following symbols are contained in the name of output node, conversion may not be successful.

    • | ; , ‘

• PRIORBOX_NODE: Node equivalent to “priorbox”

  • Need to set when IS_SSD=1

• CAVALRY_VER: Version of cavalry to use

• PREPRO: Path of preprocessing script (python script)

  • Refer to “/home/cvtool/ common/prepro.py” for how to create a script

• PREPRO_ARG: Argument of preprocessing script (python script)

• PARSER_IN_DATA_FORMAT: Format of input data for target model (when tfparser run) (NHWC or NCHW)

• IN_DATA_FILEFORMAT: Input data format

  • Examples : uint8->0,0,0,0, float32->1,2,0,7, float16-> 1,1,0,4

  • When the value of IN_DATA_FILEFORMAT changes from “0,0,0,0”, setting PREPRO is needed.

• N_DATA_TRANSPOSE: Specify when performing TRANSPOSE on the input data

Convert sample model

inception_v4

Download sample model.

$ cd /home/cvtool/conversion/tensorflow/inception_v4
$ wget https://storage.googleapis.com/download.tensorflow.org/models/tflite/model_zoo/upload_20180427/inception_v4_2018_04_27.tgz
$ tar -xvf inception_v4_2018_04_27.tgz
$ mv inception_v4.pb sample/model
$ rm inception_v4_2018_04_27.tgz

Convert the model.

$ ./tf_conversion.sh setting.conf

mobilenetv2ssd

Download sample model.

$ cd /home/cvtool/conversion/tensorflow/mobilenetv2ssd
$ wget http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz
$ tar -xvzf ssd_mobilenet_v2_coco_2018_03_29.tar.gz
$ mv ssd_mobilenet_v2_coco_2018_03_29/frozen_inference_graph.pb sample/model
$ rm ssd_mobilenet_v2_coco_2018_03_29.tar.gz

Convert the model.

$ ./tf_conversion.sh setting_coco20180329.conf