• State Resolved
  • Locked Locked
  • Replies 6 replies
  • Subscribers 100 subscribers
  • Views 479 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

TDA4VM: maxpooling problem

user6406170
Intellectual 375 points
Part Number: TDA4VM

Dear TI，

when I use a maxpooling with kernel size = 3X3 , stride = 1, padding = 1, and the maxpooling's input is all negative, but it's output is all zero.why?

  • 0
    TI__Mastermind 22145 points

    Only border outputs are expected to be zero.

    Can you share the SDK version and Model that you are using to reproduce the issue

  • 0 in reply to kumar.desappan
    Intellectual 375 points

    the version is psdk_rtos_auto_j7_07_00_00_11,and my onnx model is here. The maxpooling's output is all zero.

  • 0 in reply to user6406170
    Intellectual 375 points

    This is my onnx model file.efficientdet.zip

    And this is my tidl imported model.efficientdet_tidl.zip

    This is my import config file.

    Fullscreen tidl_import_efficientdet.txt Download 
    modelType = 2
numFrames = 1
numParamBits  = 16
numFeatureBits = 16
quantizationStyle = 3
inputNetFile       = "../../test/testvecs/models/public/onnx/efficientdet/efficientdet.onnx"
outputNetFile      = "../../test/testvecs/config/tidl_models/onnx/efficientdet.bin"
outputParamsFile   = "../../test/testvecs/config/tidl_models/onnx/efficientdet_"
inWidth  = 1024
inHeight = 512
inNumChannels = 3
inDataFormat = 0
inData = ../../test/testvecs/config/1024.txt
postProcType = 0
inDataNorm = 1
inMean = 0 0 0
inScale =  1 1 1
debugTraceLevel = 1
writeTraceLevel = 3
foldPreBnConv2D = 0
foldBnInConv2D = 1
outDataNamesList = "963,962,959,output"

  • 0 in reply to kumar.desappan
    Intellectual 375 points

    thanks, that's my own code problem.

  • 0 in reply to kumar.desappan
    Intellectual 375 points

    But the maxpooling method is different from pytorch. And when the input has many negative inputs and has padding, the output also has some problems. It will generate many zeros.   

  • 0 in reply to user6406170
    TI__Guru 56820 points

    Hi,
        Is this issue still open?

    Regards,

    Anshu