The input size of the model is wrong?

Blog
,
#1

Question/Issue:
Got the shape of data does not match the shape of input layer error code when excution in Stm32F405RGT6 MCU . After trace the source code, it seems the input feature is correct and the input size of model is wrong.
Project ID:
503620

Context/Use case:
Got the shape of data does not match the shape of input layer error code when excution in Stm32F405RGT6 MCU .

Summary:
I have a voice recognition project. I pass all the step and the result is good. Even deploy to the MCU, the compilation is nice. After runing the firmware I got the error codce -24 and found it is the shape size issue. However, after trace the source step by step, it seems the input size of model is wrong. I got the size 650 from raw feature data copied from the feature in live classfication. And got 26 feature size after MFCC DSP processing. But I got the model input size 650. These size and processing is all from the debug steps. I found the size of the model input stored in tensorData array. I manualy chage 650 to 26 and the error code becomes 0. Sure, the 4 output is not good.
I have try to check any parameter in the classifier step. I don’t if any parameter may cause the issue.

Please help me. Thank you.

Steps to Reproduce:

  1. [Step 1]
  2. [Step 2]
  3. [Step 3]

Expected Results:
[Describe what you expected to happen]

Actual Results:
[Describe what actually happened]

Reproducibility:

  • [V] Always
  • [ ] Sometimes
  • [ ] Rarely

Environment:

  • Platform: self made stm32f405rgt6 MCU
  • Build Environment Details: stm32CubeIDE 1.16.0
  • OS Version: Windows 10
  • Edge Impulse Version (Firmware): 1.56.10
    To find out Edge Impulse Version:
  • if you have pre-compiled firmware: run edge-impulse-run-impulse --raw and type AT+INFO. Look for Edge Impulse version in the output.
  • if you have a library deployment: inside the unarchived deployment, open model-parameters/model_metadata.h and look for EI_STUDIO_VERSION_MAJOR, EI_STUDIO_VERSION_MINOR, EI_STUDIO_VERSION_PATCH
  • Edge Impulse CLI Version: [e.g., 1.5.0]
  • Project Version: [e.g., 1.0.0]
  • Custom Blocks / Impulse Configuration: no custom blocks
    Logs/Attachments:
    [Include any logs or screenshots that may help in diagnosing the issue]
    新點陣圖影像
    新點陣圖影像1920×1025 319 KB

    Logs/Attachments:
    [Include any logs or screenshots that may help in diagnosing the issue]

Additional Information:
[Any other information that might be relevant to the issue]

#2

I rebuild a project ( code:506979) and use default MFCC, classifier etc.
And the result is all the same.

#3

Finally I found the solution. I am confused betwen the raw feature data and precessed feature data. I copied the precessed feature data just below the raw feature data in the live classification page.
I still got the weak result. The maximum classification value for all the output are less than 50%. I have to found the solution.

#4

Hi @hoppertw

Please detail the solution you have found so we can help to clarify what has happened.

Likely you deleted a block in the impulse design page and added it again? Occasionally a checkbox can be missed and lead to unprocessed features being fed. Guessing you hit this scenario, but please confirm :smiley:

Best

Eoin