openvino-model-api 0.2.0

Model API: model wrappers and pipelines for inference with OpenVINO

Python* Model API package

Model API package is a set of wrapper classes for particular tasks and model architectures, simplifying data preprocess and postprocess as well as routine procedures (model loading, asynchronous execution, etc...) An application feeds model class with input data, then the model returns postprocessed output data in user-friendly format.

Package structure

The Model API consists of 3 libraries:

• adapters implements a common interface to allow Model API wrappers usage with different executors. See Model API adapters section
• models implements wrappers for Open Model Zoo models. See Model API Wrappers section
• pipelines implements pipelines for model inference and manage the synchronous/asynchronous execution. See Model API Pipelines section

Prerequisites

The package requires

• one of OpenVINO supported Python version (see OpenVINO documentation for the details)
• OpenVINO™ toolkit

If you build Model API package from source, you should install the OpenVINO™ toolkit. See the options:

Use installation package for Intel® Distribution of OpenVINO™ toolkit or build the open-source version available in the OpenVINO GitHub repository using the build instructions.

Also, you can install the OpenVINO Python* package via the command:

pip install openvino

Installing Python* Model API package

Use the following command to install Model API from source:

pip install <omz_dir>/demos/common/python

Alternatively, you can generate the package using a wheel. Follow the steps below:

1. Build the wheel.

python <omz_dir>/demos/common/python/setup.py bdist_wheel

The wheel should appear in the dist folder. Name example: openmodelzoo_modelapi-0.0.0-py3-none-any.whl

2. Install the package in the clean environment with --force-reinstall key.

pip install openmodelzoo_modelapi-0.0.0-py3-none-any.whl --force-reinstall

To verify the package is installed, you might use the following command:

python -c "from openvino.model_zoo import model_api"

Model API Wrappers

The Model API package provides model wrappers, which implement standardized preprocessing/postprocessing functions per "task type" and incapsulate model-specific logic for usage of different models in a unified manner inside the application.

The following tasks can be solved with wrappers usage:

Task type	Model API wrappers
Background Matting	VideoBackgroundMatting ImageMattingWithBackground PortraitBackgroundMatting
Classification	ClassificationModel
Deblurring	Deblurring
Human Pose Estimation	HpeAssociativeEmbedding OpenPose
Instance Segmentation	MaskRCNNModel YolactModel
Monocular Depth Estimation	MonoDepthModel
Named Entity Recognition	BertNamedEntityRecognition
Object Detection	CenterNet DETR CTPN FaceBoxes NanoDet NanoDetPlus RetinaFace RetinaFacePyTorch SSD UltraLightweightFaceDetection YOLO YoloV3ONNX YoloV4 YOLOF YOLOX
Question Answering	BertQuestionAnswering
Salient Object Detection	SalientObjectDetectionModel
Semantic Segmentation	SegmentationModel

Model API Adapters

Model API wrappers are executor-agnostic, meaning it does not implement the specific model inference or model loading, instead it can be used with different executors having the implementation of common interface methods in adapter class respectively.

Currently, OpenvinoAdapter and OVMSAdapter are supported.

OpenVINO Adapter

OpenvinoAdapter hides the OpenVINO™ toolkit API, which allows Model API wrappers launching with models represented in Intermediate Representation (IR) format. It accepts a path to either xml model file or onnx model file.

OpenVINO Model Server Adapter

OVMSAdapter hides the OpenVINO Model Server python client API, which allows Model API wrappers launching with models served by OVMS.

Refer to OVMSAdapter to learn about running demos with OVMS.

For using OpenVINO Model Server Adapter you need to install the package with extra module:

pip install <omz_dir>/demos/common/python[ovms]

ONNXRuntime Adapter

ONNXRuntimeAdapter hides the ONNXRuntime, which Model API wrappers launching with models represented in ONNX format. It accepts a path to onnx file. This adapter's functionality is limited: it doesn't support model reshaping, asynchronous inference and was tested only on limited scope of models. Supported model wrappers: SSD, MaskRCNNModel, SegmentationModel, and ClassificationModel.

To use this adapter, install extra dependencies:

pip install onnx onnxruntime

Model API Pipelines

Model API Pipelines represent the high-level wrappers upon the input data and accessing model results management. They perform the data submission for model inference, verification of inference status, whether the result is ready or not, and results accessing.

The AsyncPipeline is available, which handles the asynchronous execution of a single model.

Ready-to-use Model API solutions

To apply Model API wrappers in custom applications, learn the provided example of common scenario of how to use Model API.

In the example, the SSD architecture is used to predict bounding boxes on input image "sample.png". The model execution is produced by OpenvinoAdapter, therefore we submit the path to the model's xml file.

Once the SSD model wrapper instance is created, we get the predictions by the model in one line: ssd_model(input_data) - the wrapper performs the preprocess method, synchronous inference on OpenVINO™ toolkit side and postprocess method.

import cv2
# import model wrapper class
from model_api.models import SSD
# import inference adapter and helper for runtime setup
from model_api.adapters import OpenvinoAdapter, create_core


# read input image using opencv
input_data = cv2.imread("sample.png")

# define the path to mobilenet-ssd model in IR format
model_path = "public/mobilenet-ssd/FP32/mobilenet-ssd.xml"

# create adapter for OpenVINO™ runtime, pass the model path
inference_adapter = OpenvinoAdapter(create_core(), model_path, device="CPU")

# create model API wrapper for SSD architecture
# preload=True loads the model on CPU inside the adapter
ssd_model = SSD(inference_adapter, preload=True)

# apply input preprocessing, sync inference, model output postprocessing
results = ssd_model(input_data)

To study the complex scenarios, refer to Open Model Zoo Python* demos, where the asynchronous inference is applied.