Hello,
We are evaluating the capabilities of the AM5729 [1] and of the TIDL [2], and for that we are starting with very simple models: a model that multiple it's input by a constant and another one that is based on the Boston Housing dataset.
Everything is working fine, including the output of the model (taking into account the effect quantisation) except that we don't have access to the quantisation factor which, for this particular models is essencial.
We looked through all of the examples and none of them need a quantisation factor. We understand that as classification problems, one is not really needed.
There's seems to be one exception though.
In the ssd_multibox example the output buffer is cast to float which indicates that a conversion from integer to float, using a quantisation factor (?), has been performed somewhere.
Is there anyway to get the quantisation factor? (The simulator does output them.)
Like in the ssd_multibox example, how can we get the output buffer to be in floats?
BTW: We could not find the source code for any of the examples. Only the binaries seem to be provided. Did we miss something or is this really not available?
Best regards
Adriano Carvalho
[1] https://beagleboard.org/AI
[3] https://git.ti.com/cgit/tidl/tidl-api/tree/examples/ssd_multibox/main.cpp