TDA4VM: TIDL normalize and user's own normalize

Hi Faw,

xueming tian said:

As I know that the normalize of tidl is executed by (x - mean) * scale, then quantized to int8

   This is right and the documentation for the same can be found here : 

https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/08_00_00_12/exports/docs/tidl_j7_08_00_00_10/ti_dl/docs/user_guide_html/md_tidl_fsg_steps_to_debug_mismatch.html#did_tidl_debug_pre_processing_mismtach

xueming tian said:

Is there a problem with the calculation formula? or the TIDL inner normalize have the extra process?

   I didnt follow both the questions clearly but let me explain what happens next. If foldPreBnConv2D = 1 ( default behavior) then this normalization operation will get merged to corresponding convolution layer. If foldPreBnConv2D = 0 then there will be a separate Batch Norm layer which will do this operation.

Regards,

Anshu