Image to tensorπŸ”—

Takes in an image and outputs a dense tensor in a predefined format.

InputsπŸ”—

  • image: Any image.

  • layout: Layout of output tensor.

  • outputType: The data type of output tensor elements.

  • normalizeIntensity: A Boolean flag that enables or disables intensity range normalization. If enabled, the intensity levels of the input image are scaled by equation β€œout = ((in / channel_max) * max - offset) / divisor” before they are copied to the output tensor. This is done separately for each color channel. See normalizationFactors.

  • normalizationFactors: A 3-by-3 matrix that contains the max, offset and divisor values for each color channel. Only the first row is used with grayscale images.

OutputsπŸ”—

  • tensor: A tensor in the defined format. If the input is an RGB image, the output tensor will have three elements in the β€œc” dimension. Gray images produce just one. Use color converting to convert gray levels to RGB if needed. The β€œw” and β€œh” dimensions will match the size of the input image. Use image scaling or cropping as a preprocessing tool to fix the tensor size if required by the application. In NCHW and NHWC layouts, the batch size (β€œn”) will be one.

enum LayoutπŸ”—

Supported data layouts.

Values:

enumerator ChwLayoutπŸ”—

Channels, height, width.

enumerator HwcLayoutπŸ”—

Height, width, channels.

enumerator NchwLayoutπŸ”—

Batch, channels, height, width.

enumerator NhwcLayoutπŸ”—

Batch, height, width, channels.

enum OutputTypeπŸ”—

Supported tensor data types.

Values:

enumerator FloatπŸ”—
enumerator Float16πŸ”—
enumerator DoubleπŸ”—
enumerator Uint8πŸ”—
enumerator Int8πŸ”—
enumerator Uint16πŸ”—
enumerator Int16πŸ”—
enumerator Uint32πŸ”—
enumerator Int32πŸ”—
enumerator Uint64πŸ”—
enumerator Int64πŸ”—
enumerator BooleanπŸ”—
enumerator FloatComplexπŸ”—
enumerator DoubleComplexπŸ”—