• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Subscribers 98 subscribers
  • Views 1639 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.

TIDL tidlmodelimport question

Ahan Tseng
Prodigy 190 points

Hi,

I have question about producing TIDL Network File.

Below is my prototxt file,

layer {
  name: "conv14_mbox_deconf_BBOX_anchor0_3"
  type: "Deconvolution"
  bottom: "conv14_2_mbox_conf_BBOX_anchor0_3"
  top: "conv14_mbox_deconf_BBOX_anchor0_3"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 19
    bias_term: false
    pad: 0
    group: 19
    weight_filler {
      type: "constant"
      value: 1.0
    }
    kernel_h: 1
    kernel_w: 2
    stride_h: 1
    stride_w: 2
  }
}

Error Issue:

[libprotobuf FATAL c:\users\user\desktop\otobrite_ahan\otodas\otodl_src_20171101\modules\tidl_utils\protobuf-3.2.0rc2\src\google\protobuf\repeated_field.h:1084] CHECK failed: (index) < (current_size_):

How can I solve this error ?

Ahan

  • 0
    TI__Mastermind 22145 points

    Ahan, 

    This error is raised by protobuf library. This happens when the prototxt files has some field which are not recognized by the proto file. Please check your prototxt file for this. If you could not find any missing field,Please attach the complete prototxt, we will try to re-produce the issue.

    Regrads,

    Kumar.D

  • 0 in reply to kumar.desappan
    Prodigy 190 points

    About this error,

    I found that it happened when I execute to the function TIDL_importDeconvParams(),

    About my prototxt file, the deconvolution layer, the convolution_param only have kernel_w and kernel_h, but none kernel_size.

    If I changed my code to 

    pOrgTIDLNetStructure->TIDLPCLayers[layerIndex].layerParams.convParams.kernelW        = netStructure.layer(i).convolution_param().kernel_w();

    pOrgTIDLNetStructure->TIDLPCLayers[layerIndex].layerParams.convParams.kernelH        = netStructure.layer(i).convolution_param().kernel_h();

    , the error is gone.

    So is there any possible side effect by changing the two lines of codes above? Can TIDL work with non-square kernel too?

    Ahan

  • 0 in reply to Ahan Tseng
    TI__Mastermind 22145 points
    Hi Ahan,
    We have not validated the deconvolution with this configuration. Also the stride_w and stride_h shall be 2 for deconvolution.
    Regards,
    Kumar.D