Hello,
As a result of running the tidlModelImport tool, one obtains a log file. I obtained these statements at the end of the file:
Total Giga Macs : 0.42
Total Giga Macs : 6.3 @15 fps
Total Giga Macs : 12.6 @30 fps
I have some questions regarding reading the info in the log file:
1) The number of MAC operations for this neural network is 420 Mega MAC operations for the processing of a single frame.
Is this measure independent from the core(EVE,DSP,etc)?
The tidlStatsTool parameter was "..\quantStatsTool\eve_test_dl_algo.out.exe" in my case, but it gave the same result even after deleting the tidlStatsTool parameter. There is no exe for the dsp emulation and I have not managed to build the source yet.
2) According to the TIDL User Guide, the TDA2x+ can deliver up to 70 GMACs per second (on which core?). Does that mean that the above-mentioned network can handle up to 70/0.42 which is approximately 166 frames per second? I.e. is that how we should interpret the results?
3) One needs to get the output of the network (e.g. classification result) for unit testing. How to obtain this output?
4) Is there a document where the lexicon of the log file is explained?
Thanks a lot
Best,
Safwan
