This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA2EXEVM: result compare between quantStatsTool and TIDL on board

Part Number: TDA2EXEVM

Hi:

    I validate the result between quantStatsTool and TIDL on TDA2EXEVM board.The model is a classification model and the last result is softmax val.The   TDA2EXEVM usecase is file I/O,I print tidl link outbuffer.

    I test about 10 images:some image's softmax val is exact same between  quantStatsTool and TIDL on TDA2EXEVM;

    some image's softmax val is not same,but the max index is same.for example,the quantStatsTool result is "9c 99 8d 88 97 fe 92 95 79",but the result  on TDA2EXEVM is "a9 a6 9d 97 a6 ff a1 a5 99" .

     Is that right?Is the result between  quantStatsTool and TIDL on TDA2EXEVM same byte by byte,or not?

Thanks and regards

  • The output from quantStatsTool and TIDL on TDA2EXEVM shall be same if the input to both is same and only one image is processed at a time.

    Refer below thread for more info
    e2e.ti.com/.../
  • Hi:
    Thanks your reply!

     I have 2 confuse.

    1.  I takes 1 image as input every time! And already set params:
         createParams.quantHistoryParam1 = 0;
         createParams.quantHistoryParam2 = 0;
         createParams.quantMargin = 0;

         Should I set these Params in two place? one in tidl_tb.c for quantStatsTool,another in tidlLink_algPlugin.c for tidl on TDA2EXEVM?

    2. after I change these 3 Params in tldi_tb.c, I generate a new quanStatsTool "eve_test_dl_algo.out.exe" as follow "3.5.2 Building source in host emulation" in <<TIDeepLearningLibrary_UserGuide>>.

       But with the new quanStatsTool,the model last result softmax val is all "0xff";And I change these 3 Param back,generate quanStatsTool "eve_test_dl_algo.out.exe" again,the model result softmax val is still all "0xff";but when I use the quanStatsTool "eve_test_dl_algo.out.exe" original packed in VIsionSDK0302,the model softmax val is correct,not all "0xff".

       I found the size of my generated "eve_test_dl_algo.out.exe" is only 1.36MB,but the quanStatsTool "eve_test_dl_algo.out.exe" original packed in VIsionSDK0302 size is 7.04MB.

      I don't install OpenCV3.1,Is this? But when build "eve_test_dl_algo.out.exe",there has not any error,and generate "eve_test_dl_algo.out.exe" successfully.

       Should something setting be setted when build quanStatsTool "eve_test_dl_algo.out.exe"?

        Or Is something wrong?

        

    Pls help me!

    Thanks and Regards

       



    Thanks and Regards

  • I guess the below thread is answerignt hsi question also
    e2e.ti.com/.../2477816