mltk.core.TfliteDepthwiseConvParams

class TfliteDepthwiseConvParams

Calculated Depthwise Convolution Parameters

See also

• DepthwiseConvParamsQuantized

• Quantization Specification

Properties

depth_multiplier	Depth multiplier
dilation_height_factor	Kernel dilation height factor
dilation_width_factor	Kernel dilation width factor
input_offset	Input quantization offset (i.e.
output_offset	Output quantization offset (i.e.
quantized_activation_max	Fused activation max value
quantized_activation_min	Fused activation min value
stride_height	Kernel height
stride_width	Kernel width
weights_offset	Weight (aka filters) quantization offset (i.e.
padding	Kernel padding
per_channel_output_multiplier	Per layer multipliers for output scalers
per_channel_output_shift	Per layer shifts for output scalers

Methods

__init__
calculate	Calculate the parameters for the given layer

padding: TflitePadding

Kernel padding

stride_width: int = 0

Kernel width

stride_height: int = 0

Kernel height

dilation_width_factor: int = 1

Kernel dilation width factor

dilation_height_factor: int = 1

Kernel dilation height factor

depth_multiplier: int = 0

Depth multiplier

input_offset: int = 0

Input quantization offset (i.e. zero point)

weights_offset: int = 0

Weight (aka filters) quantization offset (i.e. zero point)

output_offset: int = 0

Output quantization offset (i.e. zero point)

per_channel_output_multiplier: List[int]

Per layer multipliers for output scalers

per_channel_output_shift: List[int]

Per layer shifts for output scalers

quantized_activation_min: int = 0

Fused activation min value

quantized_activation_max: int = 0

Fused activation max value

static calculate(layer)

Calculate the parameters for the given layer

Return type:

TfliteDepthwiseConvParams

__init__(padding=<factory>, stride_width=0, stride_height=0, dilation_width_factor=1, dilation_height_factor=1, depth_multiplier=0, input_offset=0, weights_offset=0, output_offset=0, per_channel_output_multiplier=<factory>, per_channel_output_shift=<factory>, quantized_activation_min=0, quantized_activation_max=0)

Parameters:

• padding (TflitePadding) –

• stride_width (int) –

• stride_height (int) –

• dilation_width_factor (int) –

• dilation_height_factor (int) –

• depth_multiplier (int) –

• input_offset (int) –

• weights_offset (int) –

• output_offset (int) –

• per_channel_output_multiplier (List[int]) –

• per_channel_output_shift (List[int]) –

• quantized_activation_min (int) –

• quantized_activation_max (int) –

Return type:

None