Normalization Configuration

Normalization is to limit the data to be processed within a certain range to facilitate subsequent data processing. Specially, AIPP normalization is implemented by subtracting the mean value and multiplying a factor. This can be used to achieve general normalization and also data reformatting.

For example, unit8 data to fp16 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:

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