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.

TDA4VMXEVM: TIDL issue - unable to load inference binary

Part Number: TDA4VMXEVM

Team,

We have a customer trying to evaluate our deep learning library, following this tutorial:  They are trying to benchmark this model with TI’s MMA engine. Can you help resolve the error described below?

From the customer: "I am able to convert the model using the TIDL toolchain and run test using the PC emulator. 
The problem I met when trying to deploy the model to the board is that after format the SD and put the testvecs folder in, to configure EVM (NO BOOT), and set up CCS, GELS, CCXML. The CCS cannot load the inference binary. Is this the correct instruction to follow in order to test the inference ?

I followed another instruction earlier with the board configured to be SD BOOT MODE with RTOS SDK package and a tidl example included and it was working fine.


I have uploaded our denseline model, test image and config file here."testvecs.zip

  • Brandon,

    Thanks for helping to post this issue. @TI team, please contact me if there's any question and keep us updated.

    -Jian S.

  • Hi,

    Do you get any error on the console window ?

    Is your SD-card a 4 GB card or below ?

    regards,

    Victor

  • Hi Victor,

    Yes I was using a 4GB card at that time. Could you help to try the package in your environment? We would like to get some performance data such as latency for further discussion with TI team.

    Kai and Brandon were helping with this issue until required to WFH so could not get access to their EVM.

    Thanks,

    Jian S.

  • Hi Jian,

    Here is the output I obtained on the EVM:

    Processing config file #0 : testvecs/config/infer/public/caffe/tidl_infer_denseline.txt
    Syncd
    ----------------------- TIDL Process with TARGET DATA FLOW ------------------------

    # 0jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.43 ... .... .....
    # 1jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.29 ... .... .....
    # 2jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.27 ... .... .....
    # 3jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.28 ... .... .....
    # 4jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.29 ... .... .....
    # 5jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.28 ... .... .....
    # 6jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.29 ... .... .....
    # 7jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.28 ... .... .....
    # 8jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.29 ... .... .....
    # 9jpg file read not supported. Only BMP file read is supported in target
    . .. TSC Mega Cycles = 25.29 ... .... .....

    regards,

    Victor

  • Hi Victor,

    Thanks for the experiment. Is BMP file the only format supported in target? I provide inFileFormat = 2 in tidl_infer_denseline.txt which is a list of jpg files.

    Thanks,

    Jian

  • Hi Jian,

    If you have set inFileFormat to 2 in the inference config file then only bmp is supported on the target. Otherwise on PC, jpg/png/bmp are supported.

    regards,

    Victor

  • Hi Victor,

    I see, thanks for the clarification. I changed image format, could you help to try this again? I do not have access to EVM at this moment.

    And according to your result, is the inference time ~25ms? 

    Thanks

    -Jian 

    7585.testvecs.zip

  • Hi Jian,

    Yes 25 ms is the inference time.

    Here is my output with your new files:

    Processing config file #0 : testvecs/config/infer/public/caffe/tidl_infer_denseline.txt
    Syncd
    ----------------------- TIDL Process with TARGET DATA FLOW ------------------------

    # 0 . .. TSC Mega Cycles = 25.42 ... .... .....
    # 1 . .. TSC Mega Cycles = 25.42 ... .... .....
    # 2 . .. TSC Mega Cycles = 25.39 ... .... .....
    # 3 . .. TSC Mega Cycles = 25.43 ... .... .....
    # 4 . .. TSC Mega Cycles = 25.42 ... .... .....
    # 5 . .. TSC Mega Cycles = 25.41 ... .... .....
    # 6 . .. TSC Mega Cycles = 25.41 ... .... .....
    # 7 . .. TSC Mega Cycles = 25.41 ... .... .....
    # 8 . .. TSC Mega Cycles = 25.41 ... .... .....
    # 9 . .. TSC Mega Cycles = 25.40 ... .... .....

  • I am also attaching the output of the model.denseline.bin