Interface TypedTensor<T>

Represent multi-dimensional arrays to feed to or fetch from model inferencing.

interface TypedTensor<T> {
    data: DataTypeMap[T];
    dims: readonly number[];
    gpuBuffer: GpuBufferTypeFallback;
    location: DataLocation;
    mlTensor: MLTensorTypeFallback;
    size: number;
    texture: WebGLTexture;
    type: T;
    dispose(): void;
    getData(releaseData?): Promise<DataTypeMap[T]>;
    reshape(dims): TypedTensor<T>;
    toDataURL(options?): string;
    toImageData(options?): ImageData;
}

Type Parameters

Hierarchy

  • TypedTensorBase<T>
  • TypedTensorUtils<T>
    • TypedTensor

Properties

data dims gpuBuffer location mlTensor size texture type

Methods

dispose getData reshape toDataURL toImageData

Properties

Readonly data

data: DataTypeMap[T]

Get the buffer data of the tensor.

If the data is not on CPU (eg. it's in the form of WebGL texture or WebGPU buffer), throw error.

Readonly dims

dims: readonly number[]

Get the dimensions of the tensor.

Readonly gpuBuffer

gpuBuffer: GpuBufferTypeFallback

Get the WebGPU buffer that holds the tensor data.

If the data is not on GPU as WebGPU buffer, throw error.

Readonly location

location: DataLocation

Get the location of the data.

Readonly mlTensor

mlTensor: MLTensorTypeFallback

Get the WebNN MLTensor that holds the tensor data.

If the data is not in a WebNN MLTensor, throw error.

Readonly size

size: number

Get the number of elements in the tensor.

Readonly texture

texture: WebGLTexture

Get the WebGL texture that holds the tensor data.

If the data is not on GPU as WebGL texture, throw error.

Readonly type

type: T

Get the data type of the tensor.

Methods

dispose

  • dispose(): void

  • Dispose the tensor data.

    If the data is on CPU, remove its internal reference to the underlying data. If the data is on GPU, release the data on GPU.

    After calling this function, the tensor is considered no longer valid. Its location will be set to 'none'.

    Returns void

getData

  • getData(releaseData?): Promise<DataTypeMap[T]>

  • Get the buffer data of the tensor.

    If the data is on CPU, returns the data immediately. If the data is on GPU, downloads the data and returns the promise.

    Parameters

    • Optional releaseData: boolean

      whether release the data on GPU. Ignore if data is already on CPU.

    Returns Promise<DataTypeMap[T]>

reshape

  • reshape(dims): TypedTensor<T>

  • Create a new tensor with the same data buffer and specified dims.

    Parameters

    • dims: readonly number[]

      New dimensions. Size should match the old one.

    Returns TypedTensor<T>

toDataURL

  • toDataURL(options?): string

  • creates a DataURL instance from tensor

    Parameters

    • Optional options: TensorToDataUrlOptions

      An optional object representing options for creating a DataURL instance from the tensor.

      The following default settings will be applied:

      • format: 'RGB'
      • tensorLayout: 'NCHW'

    Returns string

    a DataURL string representing the image converted from tensor data

toImageData

  • toImageData(options?): ImageData

  • creates an ImageData instance from tensor

    Parameters

    • Optional options: TensorToImageDataOptions

      An optional object representing options for creating an ImageData instance from the tensor.

      The following default settings will be applied:

      • format: 'RGB'
      • tensorLayout: 'NCHW'

    Returns ImageData

    an ImageData instance representing the image converted from tensor data

Settings

Member Visibility

Theme

On This Page

Generated using TypeDoc