Encoder for visibility weights. More...

#include <weightencoder.h>

Public Types
typedef T	value_t
	Value type of the original weights (a floating point value). More...

typedef unsigned	symbol_t
	Value type of the symbols to which the weights are encoded. More...

Public Member Functions
	WeightEncoder (unsigned quantCount)
	Construct a new encoder with the given quantization count. More...

void	Encode (value_t &scaleVal, std::vector< symbol_t > &dest, const std::vector< value_t > &input) const
	Encodes a vector of values. More...

void	Decode (std::vector< value_t > &dest, value_t scaleVal, const std::vector< symbol_t > &input) const
	Decode a previously encoded value. More...

Private Attributes
unsigned	_quantCount

Detailed Description

template<typename T>
class dyscostman::WeightEncoder< T >

Encoder for visibility weights.

It's a linear quantizer for non-negative values, with a single scaling factor. The scaling factor will be such that the max value will still fit.

Author: André Offringa

Definition at line 16 of file weightencoder.h.

Member Typedef Documentation

template<typename T >

typedef unsigned dyscostman::WeightEncoder< T >::symbol_t

Value type of the symbols to which the weights are encoded.

Definition at line 21 of file weightencoder.h.

template<typename T >

typedef T dyscostman::WeightEncoder< T >::value_t

Value type of the original weights (a floating point value).

Definition at line 19 of file weightencoder.h.

Constructor & Destructor Documentation

template<typename T >

dyscostman::WeightEncoder< T >::WeightEncoder ( unsigned quantCount )

inlineexplicit

Construct a new encoder with the given quantization count.

Parameters

quantCount The number of quantization levels.

Definition at line 26 of file weightencoder.h.

Member Function Documentation

template<typename T >

void dyscostman::WeightEncoder< T >::Decode	(	std::vector< value_t > &	dest,
		value_t	scaleVal,
		const std::vector< symbol_t > &	input
	)		const

inline

Decode a previously encoded value.

Parameters

dest	The destination for the decoded weight values.
scaleVal	The scale value for the vector of values.
input	The input symbols to be decoded.

See Also: Encode()

Definition at line 66 of file weightencoder.h.

References dyscostman::WeightEncoder< T >::_quantCount.

template<typename T >

void dyscostman::WeightEncoder< T >::Encode	(	value_t &	scaleVal,
		std::vector< symbol_t > &	dest,
		const std::vector< value_t > &	input
	)		const

inline

Encodes a vector of values.

Will return a vector of symbols and a scaling value. Together, these can be used to get the original values back (with some loss due to quantization), by using Decode().

Parameters

scaleVal	Will receive the scale value for the vector of values
dest	The destinal vector with symbols. It is assumed that it is already of the right size, i.e., equal to input.size().
input	The input array of values to be encoded.

Find max value (implicit assumption input is not empty)

Definition at line 37 of file weightencoder.h.

References dyscostman::WeightEncoder< T >::_quantCount.