Normalization Configuration

Normalization is a processing method that scales data to a specific range, facilitating subsequent data processing and analysis. The normalization function supported by AIPP is implemented through two steps: subtracting the mean value and multiplying a factor. This processing method not only meets common normalization requirements, but also supports flexible conversion between multiple data formats, improving data processing efficiency and compatibility.

For example, unit8-to-fp16 data conversion can be represented by the following formula, where mean_chn_i indicates the mean value per channel, min_chn_i the minimum value per channel, and var_reci_chn_i the reciprocal of the variance per channel. These three parameters need to be configured for each channel.

pixel_out_chx(i)=[pixel_in_chx(i)-mean_chn_i-min_chn_i]*var_reci_chn_i