[JS OV] Node.js API

[vishniakov-nikolai]

What's included

• Node.js bindings for OpenVINO
• Samples & notebooks

Details

Folders structure

/samples/js/
  ./node/
    ./notebooks/ # VSCode typescript notebooks
    ./hello_classification/
    ./hello_reshape_ssd/
    ./classification_sample_async/
/src/bindings/js/
  ./docs/
  ./node/
    ./include/ # bindings header files
    ./lib/     # typescript sources of js part
    ./src/     # c++ bindings sources
    ./tests/   # js unit-tests

JS API

Exposed entities:

{
  addon: {
    Core: CoreConstructor,
    Tensor: TensorConstructor,
    PartialShape: PartialShapeContructor,

    element: typeof element,
    preprocess: {
      resizeAlgorithm: typeof resizeAlgorithm,
      PrePostProcessor: PrePostProcessorConstructor,
    }
  }
}

Detailed description:

Core {
  constructor();
  compileModel(model: Model, device: string, config?: { [option: string]: string }): Promise<CompiledModel>;
  compileModelSync(model: Model, device: string, config?: { [option: string]: string }): CompiledModel;
  readModel(modelPath: string, weightsPath?: string): Promise<Model>;
  readModelSync(modelPath: string, weightsPath?: string): Model;
  readModel(modelBuffer: Uint8Array, weightsBuffer?: Uint8Array): Promise<Model>;
  readModelSync(modelBuffer: Uint8Array, weightsBuffer?: Uint8Array): Model;
}

Model {
  outputs: Output[];
  inputs: Output[];
  output(nameOrId?: string | number): Output;
  input(nameOrId?: string | number): Output;
  getName(): string;
}

CompiledModel {
  outputs: Output[];
  inputs: Output[];
  output(nameOrId?: string | number): Output;
  input(nameOrId?: string | number): Output;
  createInferRequest(): InferRequest;
}

Tensor {
  data: number[];
  constructor(type: element,
      shape: number[],
      tensorData?: number[] | SupportedTypedArray);
  getElementType(): element;
  getShape(): number[];
  getData(): number[];
}

InferRequest {
  setTensor(name: string, tensor: Tensor): void;
  setInputTensor(idxOrTensor: number | Tensor, tensor?: Tensor): void;
  setOutputTensor(idxOrTensor: number | Tensor, tensor?: Tensor): void;
  getTensor(nameOrOutput: string | Output): Tensor;
  getInputTensor(idx?: number): Tensor;
  getOutputTensor(idx?: number): Tensor;
  getCompiledModel(): CompiledModel;
  infer(inputData?: { [inputName: string]: Tensor | SupportedTypedArray} | Tensor[] | SupportedTypedArray[]): 
    { [outputName: string] : Tensor };
  inferAsync(inputData: { [inputName: string]: Tensor} | Tensor[] ): Promise<{ [outputName: string] : Tensor}>;
}

Output {
  anyName: string;
  shape: number[];
  toString(): string;
  getAnyName(): string;
  getShape(): number[];
  getPartialShape(): PartialShape;
}

PrePostProcessor {
  constructor(model: Model);
  build(): PrePostProcessor;
  input(idxOrTensorName?: number | string): InputInfo;
  output(idxOrTensorName?: number | string): OutputInfo;
}

InputTensorInfo {
  setElementType(elementType: element | elementTypeString ): InputTensorInfo;
  setLayout(layout: string): InputTensorInfo;
  setShape(shape: number[]): InputTensorInfo;
}

OutputTensorInfo {
  setElementType(elementType: element | elementTypeString ): InputTensorInfo;
  setLayout(layout: string): InputTensorInfo;
}

PreProcessSteps {
  resize(algorithm: resizeAlgorithm | string): PreProcessSteps;
}

InputModelInfo {
  setLayout(layout: string): InputModelInfo;
}

InputInfo {
  tensor(): InputTensorInfo;
  preprocess(): PreProcessSteps;
  model(): InputModelInfo;
}

OutputInfo {
  tensor(): OutputTensorInfo;
}

PartialShape {
  constructor(shape: string);
  isStatic(): boolean;
  isDynamic(): boolean;
  toString(): string;
  getDimensions(): Dimension[];
}

enum element {
  u8,
  u32,
  u16,
  i8,
  i16,
  i32,
  i64,
  f32,
  f64,
}

enum resizeAlgorithm {
  RESIZE_NEAREST,
  RESIZE_CUBIC,
  RESIZE_LINEAR,
}

Tickets

• 119281

[andrei-kochin]

@vishniakov-nikolai do we need to store images in repo instead of downloading it from public source?

[vishniakov-nikolai]

@vishniakov-nikolai do we need to store images in repo instead of downloading it from public source?

Of course it isn't needed. I'll make it differently.

[ilya-lavrenov]

build_jenkins

[ilya-lavrenov]

build_jenkins

[mlukasze]

build_jenkins

[vishniakov-nikolai]

build_jenkins

[vishniakov-nikolai]

build_jenkins

[akuporos]

build_jenkins

[vishniakov-nikolai]

build_jenkins