TDA4VM: Different results while inferencing in TIDL and on device.

Due to regional holiday, the expert handling this thread  is currently out of office. Please expect a 1~2 day delay in responses.

Apologies for the delay, and thank you for you patience.

Hi,

Which sdk version you are using ?

Are you using OSRT or TIDL RT ? 

Also is it possible to share the model compilation logs for initial analysis?

How are you comparing the results ? fixed pointed converted to float and diff with reference float model values ? What bit quantized is your model (8 / 16)

May i know what percentage of offset you are experiencing ? 

You can refer to our doc on model accuracy here : https://github.com/TexasInstruments/edgeai-tidl-tools/blob/master/docs/tidl_osr_debug.md#steps-to-debug-functional-mismatch-in-host-emulation