TDA4VH-Q1: How to compare layer by layer without c7x offload vs with c7x offload

Hi,

Thanks for question, one of our analytics expert will look into this.

Thanks

Hi, 

I apologize for the delay in getting to your question, as I was out of office and am now catching up.

To clarify, are these graphs comparing ARM execution with float32 versus C7x offload (and in this case you've set tensor_bits to 32)? 

Best,

Asha

Hi,

Yes, and to clarify your C7x offload is with what set as tensor_bits? Comparing layers is documented here (https://github.com/TexasInstruments/edgeai-tidl-tools/blob/master/docs/tidl_osr_debug.md#feature-map-comparison-with-reference) if that is what you are looking for. 

Would you also be able to provide the overlapping pixel heatmap, so we can look at that graph view as well? Is there also a specific accuracy level that you are targeting?

I am also assuming this is the same model you have shared with us via email through your FAE so I can look at this on my end?

Best,

Asha