blobconverter 0.0.3

The tool that allows you to covert neural networks to MyriadX blob

BlobConverter CLI

This tool allows you to convert neural network files (from various sources, like Tensorflow, Caffe or OpenVINO) to MyriadX blob file

Installation

python3 -m pip install blobconverter

Usage

usage: blobconverter [-h] [-zn ZOO_NAME] [-cp CAFFE_PROTO] [-cm CAFFE_MODEL] [-tf TENSORFLOW_PB] [-ox OPENVINO_XML] [-ob OPENVINO_BIN]
                     [-rawn RAW_NAME] [-rawc RAW_CONFIG] [-sh {1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16}] [-dt DATA_TYPE] [-o OUTPUT_DIR]
                     [-v VERSION] [--optimizer-params OPTIMIZER_PARAMS] [--compile-params COMPILE_PARAMS] [--converter-url URL] [--no-cache]

optional arguments:
  -h, --help            show this help message and exit
  -zn ZOO_NAME, --zoo-name ZOO_NAME
                        Name of a model to download from OpenVINO Model Zoo
  -cp CAFFE_PROTO, --caffe-proto CAFFE_PROTO
                        Path to Caffe .prototxt file
  -cm CAFFE_MODEL, --caffe-model CAFFE_MODEL
                        Path to Caffe .caffemodel file
  -tf TENSORFLOW_PB, --tensorflow-pb TENSORFLOW_PB
                        Path to TensorFlow .pb file
  -ox OPENVINO_XML, --openvino-xml OPENVINO_XML
                        Path to OpenVINO .xml file
  -ob OPENVINO_BIN, --openvino-bin OPENVINO_BIN
                        Path to OpenVINO .bin file
  -rawn RAW_NAME, --raw-name RAW_NAME
                        Name of the converted model (advanced)
  -rawc RAW_CONFIG, --raw-config RAW_CONFIG
                        Path to raw .yml file with model config (advanced)
  -sh {1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16}, --shaves {1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16}
                        Specifies number of SHAVE cores that converted model will use
  -dt DATA_TYPE, --data-type DATA_TYPE
                        Specifies model data type
  -o OUTPUT_DIR, --output-dir OUTPUT_DIR
                        Directory where the output blob should be saved
  -v VERSION, --version VERSION
                        OpenVINO version to use for conversion
  --optimizer-params OPTIMIZER_PARAMS
                        Additional params to use when converting a model to OpenVINO IR
  --compile-params COMPILE_PARAMS
                        Additional params to use when compiling a model to MyriadX blob
  --converter-url URL   URL to BlobConverter API endpoint used for conversion
  --no-cache            Omit .cache directory and force new compilation of the blob

Conversion examples (cli)

OpenVINO Model Zoo

python3 -m blobconverter --zoo-name face-detection-retail-0004 --shaves 6

Caffe

python3 -m blobconverter --caffe-proto /path/to/mobilenet-ssd.prototxt --caffe-model /path/to/mobilenet-ssd.caffemodel --shaves 6

TensorFlow

python3 -m blobconverter --tensorflow-pb /path/to/deeplabv3_mnv2_pascal_train_aug.pb --optimizer-params --reverse_input_channels --input_shape=[1,513,513,3] --input=1:mul_1 --output=ArgMax --shaves 6

OpenVINO IR

python3 -m blobconverter --openvino-xml /path/to/face-detection-retail-0004.xml --openvino-bin /path/to/face-detection-retail-0004.bin --shaves 7

Raw model config (advanced)

python3 -m blobconverter --raw-config /path/to/model.yml --raw-name license-plate-recognition-barrier-0007 --shaves 6

Conversion examples (Python)

OpenVINO Model Zoo

import blobconverter

blob_path = blobconverter.from_zoo(
    name="face-detection-retail-0004", 
    shaves=6,
)

Caffe

import blobconverter

blob_path = blobconverter.from_caffe(
    proto="/path/to/mobilenet-ssd.prototxt",
    model="/path/to/mobilenet-ssd.caffemodel",
    data_type="FP16",
    shaves=5,
)

TensorFlow

import blobconverter

blob_path = blobconverter.from_tf(
    frozen_pb="/path/to/deeplabv3_mnv2_pascal_train_aug.pb",
    data_type="FP16",
    shaves=5,
    optimizer_params=[
        "--reverse_input_channels",
        "--input_shape=[1,513,513,3]",
        "--input=1:mul_1",
        "--output=ArgMax",
    ],
)

OpenVINO IR

import blobconverter

blob_path = blobconverter.from_openvino(
    xml="/path/to/face-detection-retail-0004.xml",
    bin="/path/to/face-detection-retail-0004.bin",
    data_type="FP16",
    shaves=5,
)

Raw model config (advanced)

import blobconverter

blob_path = blobconverter.from_config(
    name="license-plate-recognition-barrier-0007",
    path="/path/to/model.yml",
    data_type="FP16",
    shaves=5,
)