api.md

API Documentation

Introduction

This is the API doc of inference-engine-node.

Interfaces

InferenceEngine

dictonary ApiVersion {
  long major;
  long minor;
};

dictionary Version {
  DOMString description;
  DOMString buildNumber;
  ApiVersion apiVersion;
};

interface InferenceEngine {
    Version getVersion();
};

Core

interface PluginVersions {
  readonly maplike<DOMString, Version>;
};

interface Core {
  PluginVersions getVersions(DOMString deviceName);
  sequence<DOMString> getAvailableDevices();
  Promise<Network> readNetwork(DOMString modelFilePath, [DOMString weightsFilePath]);
  Promise<Network> readNetworkFromData(DOMString model, ArrayBuffer weights);
  Promise<ExecutableNetwork> loadNetwork(Network network, DOMString deviceName);
};

Network

enum Precision {
  'unspecified',
  'mixed',
  'fp32',
  'fp16',
  'q87',
  'i16',
  'u8',
  'i8',
  'u16',
  'i32',
  'i64',
  'bin',
  'custom'
};

enum Layout {
  'any',
  'nchw',
  'nhwc',
  'ncdhw',
  'ndhwc',
  'oihw',
  'scalar',
  'c',
  'chw',
  'hw',
  'nc',
  'cn',
  'blocked'
};

enum ColorFormat {
  'raw',
  'rgb',
  'bgr',
  'rgbx',
  'bgrx',
  'nv12',
  'i420'
};

enum ResizeAlgorithm {
  'no_resize',
  'resize_bilinear',
  'resize_area',
};

enum MeanVariant {
  'mean_image',
  'mean_value',
  'none',
};


interface InputInfo {
  DOMString name();
  Precision getPrecision();
  void setPrecision(Precision precision);
  Layout getLayout();
  void setLayout(Layout layout);
  sequence<unsigned long long> getDims();
  PreProcessInfo getPreProcess()
};

interface PreProcessInfo {
  void init(unsigned long numOfChannels);
  ColorFormat getColorFormat();
  void setColorFormat(ColorFormat colorformat);
  ResizeAlgorithm getResizeAlgorithm();
  void setResizeAlgorithm(ResizeAlgorithm resizeAlgorithm);
  MeanVariant getMeanVariant();
  void setVariant(MeanVariant meanVariant);
  unsigned long getNumberOfChannels();
  PreprocessChannel getPreProcessChannel(unsigned long indexOfChannels);
};

interface TensorDesc {
  required Precision precision;
  required sequence<unsigned long long> dims,
  required Layout layout;
};

interface MeanData {
  required TensorDesc desc;
  required ArrayBuffer data;
};

interface PreProcessChannel {
  optional float stdScale = 1;
  optional float meanValue = 0;
  optional MeanData meanData = null;
};

interface OutputInfo {
  DOMString name();
  Precision getPrecision();
  void setPrecision(Precision precision);
  Layout getLayout();
  sequence<unsigned long long> getDims();
};

interface Network {
  DOMString getName();
  sequence<InputInfo> getInputsInfo();
  sequence<OutputInfo> getOutputsInfo();
};

ExecutableNetwork

interface Blob {
  unsigned long byteSize();
  unsigned long size();
  ArrayBuffer rmap();
  ArrayBuffer wmap();
  ArrayBuffer rwmap();
  void unmap();
};

interface InferRequest {
  Blob getBlob(DOMString name);
  void infer();
  Promise<void> startAsync();
};

interface ExecutableNetwork {
  InferRequest createInferRequest();  
};

Example

// ----------- 1. Load inference engine instance -------------------------------
const core = new ie.Core();
// -----------------------------------------------------------------------------

// ----------- 2. Read IR Generated by ModelOptimizer (.xml and .bin files) ----
const network = await core.readNetwork(model_xml, model_bin);
// -----------------------------------------------------------------------------

// ----------- 3. Configure input & output -------------------------------------
// -------------- Prepare input blobs ------------------------------------------
const input_info = network.getInputsInfo()[0];
input_info.setLayout('nhwc');
input_info.setPrecision('u8');

// -------------- Prepare output blobs -----------------------------------------
const output_info = network.getOutputsInfo()[0];
output_info.setPrecision('fp32');
// -----------------------------------------------------------------------------

// ----------- 4. Loading model to the device ----------------------------------
const executable_network = await core.loadNetwork(network, 'CPU');
// -----------------------------------------------------------------------------

// ----------- 5. Create infer request -----------------------------------------
const infer_request = executable_network.createInferRequest();
// -----------------------------------------------------------------------------

// ----------- 6. Prepare input ------------------------------------------------
const input_blob = infer_request.getBlob(input_info.name());
const input_data = new Uint8Array(input_blob.wmap());
// fill input_data
input_blob.unmap();
// -----------------------------------------------------------------------------

// ----------- 7. Do inference asynchronously-----------------------------------
await infer_request.startAsync();
// -----------------------------------------------------------------------------

// ----------- 8. Process output -----------------------------------------------
const output_blob = infer_request.getBlob(output_info.name());
const output_data = new Float32Array(output_blob.rmap());
// process output_data
output_blob.unmap();
// -----------------------------------------------------------------------------