Basic Geometry, Numerical Algorithms and Interfaces
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Groups Pages
List of all members
FloatCompressor Class Reference

Detailed Description

Encodes 32bit floats to 16bit, with loss of range and precision.

The IEEE-754 half-float format has a 5bit exponent and 11 bit significand. Relative accuracy is about 1e-3; the maximum representable value is 65504.

Only use this format instead of linearized quantization if logarithmic resolution is needed, that is, the relative error of represented values should remain constant for all values. Fixed-point representation is better if the absolute error is of interest.


#include <floatcompressor.h>

The documentation for this class was generated from the following file: