Structures

The following structures are available globally.

  • Describes how pixel values in the range of [0,255] will be normalized for non-quantized, float32 models.

    Pixels will typically normalized to values in the range [0,1] or [-1,+1], although separate biases may be applied to each of the RGB channels.

    Pixel normalization is like quantization but in the opposite direction.

    See more

    Declaration

    Objective-C

    struct TIOPixelNormalization {}

    Swift

    struct TIOPixelNormalization
  • Describes how floating point data will be quantized to uint8_t data.

    @field scale A scaling value. @field bias A bias term added after the scale is applied.

    Data is quantized according to the following equation: @code quantized_value = value * scale + bias @endcode

    See more

    Declaration

    Objective-C

    struct TIODataQuantization {}

    Swift

    struct TIODataQuantization
  • Describes how uint8_t data will be dequantized back into a floating point representation.

    @field scale A scaling value. @field bias A bias term added after the scale is applied.

    Data is dequantized according to the following equation: @code dequantized_value = (value + bias) * scale @endcode

    See more

    Declaration

    Objective-C

    struct TIODataDequantization {}

    Swift

    struct TIODataDequantization
  • Describes the input volume of a tensor that takes an image

    See more

    Declaration

    Objective-C

    struct TIOImageVolume {}

    Swift

    struct TIOImageVolume