TDA4VM: Importing models and running imported models on target

Damon@INVO

Part Number: TDA4VM

Hi TI,

I use TI official open source project "github.com/.../caffe-0.17" for semantic segmentation model training. The following three models are used:

jsegnet21v2
mobilesegnet-1.0
mobilesegnetv2-1.0

The above model uses scripts "https://github.com/mathmanu/caffe-jacinto-models/blob/caffe-0.17/scripts/train_mobilenet_segmentation.sh".

When TIDL toolkit is used for model transformation on PC, the three models are successfully transformed and the correct results can be output. The following are the commands executed and the output:

the left is the input image, the right is output result of the transformed model:

No problems were encountered in the above three model transformations and the output results of the model on PC. However, when the transformed model is run on TDA4 embedded device, there is a problem: only jsegnet21v2 model can get the correct prediction results, while mobilesegnet-1.0 and mobilesegnetv2-1.0 models can't output the prediction results normally. I looked at the model files of the three models and compared the supported Caffe model layer types in the documentation and found that all layers of the models were within the supported range.

The output of jsegnet21v2 model on TDA4 is normal
The output of the mobilesegnet-1.0 model on TDA4 is abnormal

The output of the mobilesegnetv2-1.0 model on TDA4 is abnormal

This problem has bothered me for a long time. I hope you can help me solve it. thanks.

Regards,

Damon

over 4 years ago

0 kumar.desappan over 4 years ago

TI__Mastermind 22145 points

Could you please share the application (Binary file location in SDK) that you are using to test these models on Device?

BTW, the steps in the below doc can be used for debugging the functional mismatches

software-dl.ti.com/.../md_tidl_fsg_steps_to_debug_mismatch.html

0 Damon@INVO over 4 years ago in reply to kumar.desappan

Intellectual 750 points

Hi kumar:

Thank you for answering my question. By 'Device', do you mean my simulation PC ?

I wonder if there is something wrong with the model itself.

This is my Caffe training network:

name: "mobilesegnett1-1.0_train"
layer {
  name: "data"
  type: "ImageLabelData"
  top: "data"
  top: "label"
  transform_param {
    mirror: true
    crop_size: 320
    mean_value: 0.0
  }
  image_label_data_param {
    image_list_path: "../data/train-image-lmdb"
    label_list_path: "../data/train-label-lmdb"
    batch_size: 8
    shuffle: true
    threads: 1
    backend: LMDB
  }
}
layer {
  name: "data/bias"
  type: "Bias"
  bottom: "data"
  top: "data/bias"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  bias_param {
    filler {
      type: "constant"
      value: -128.0
    }
  }
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data/bias"
  top: "conv1"
  convolution_param {
    num_output: 32
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 1
    stride: 2
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv1/bn"
  type: "BatchNorm"
  bottom: "conv1"
  top: "conv1"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "conv2_1/dw"
  type: "Convolution"
  bottom: "conv1"
  top: "conv2_1/dw"
  convolution_param {
    num_output: 32
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 32
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv2_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv2_1/dw"
  top: "conv2_1/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu2_1/dw"
  type: "ReLU"
  bottom: "conv2_1/dw"
  top: "conv2_1/dw"
}
layer {
  name: "conv2_1/sep"
  type: "Convolution"
  bottom: "conv2_1/dw"
  top: "conv2_1/sep"
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv2_1/sep/bn"
  type: "BatchNorm"
  bottom: "conv2_1/sep"
  top: "conv2_1/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu2_1/sep"
  type: "ReLU"
  bottom: "conv2_1/sep"
  top: "conv2_1/sep"
}
layer {
  name: "conv2_2/dw"
  type: "Convolution"
  bottom: "conv2_1/sep"
  top: "conv2_2/dw"
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 64
    stride: 2
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv2_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv2_2/dw"
  top: "conv2_2/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu2_2/dw"
  type: "ReLU"
  bottom: "conv2_2/dw"
  top: "conv2_2/dw"
}
layer {
  name: "conv2_2/sep"
  type: "Convolution"
  bottom: "conv2_2/dw"
  top: "conv2_2/sep"
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv2_2/sep/bn"
  type: "BatchNorm"
  bottom: "conv2_2/sep"
  top: "conv2_2/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu2_2/sep"
  type: "ReLU"
  bottom: "conv2_2/sep"
  top: "conv2_2/sep"
}
layer {
  name: "conv3_1/dw"
  type: "Convolution"
  bottom: "conv2_2/sep"
  top: "conv3_1/dw"
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 128
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv3_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv3_1/dw"
  top: "conv3_1/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu3_1/dw"
  type: "ReLU"
  bottom: "conv3_1/dw"
  top: "conv3_1/dw"
}
layer {
  name: "conv3_1/sep"
  type: "Convolution"
  bottom: "conv3_1/dw"
  top: "conv3_1/sep"
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv3_1/sep/bn"
  type: "BatchNorm"
  bottom: "conv3_1/sep"
  top: "conv3_1/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu3_1/sep"
  type: "ReLU"
  bottom: "conv3_1/sep"
  top: "conv3_1/sep"
}
layer {
  name: "conv3_2/dw"
  type: "Convolution"
  bottom: "conv3_1/sep"
  top: "conv3_2/dw"
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 128
    stride: 2
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv3_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv3_2/dw"
  top: "conv3_2/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu3_2/dw"
  type: "ReLU"
  bottom: "conv3_2/dw"
  top: "conv3_2/dw"
}
layer {
  name: "conv3_2/sep"
  type: "Convolution"
  bottom: "conv3_2/dw"
  top: "conv3_2/sep"
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv3_2/sep/bn"
  type: "BatchNorm"
  bottom: "conv3_2/sep"
  top: "conv3_2/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu3_2/sep"
  type: "ReLU"
  bottom: "conv3_2/sep"
  top: "conv3_2/sep"
}
layer {
  name: "conv4_1/dw"
  type: "Convolution"
  bottom: "conv3_2/sep"
  top: "conv4_1/dw"
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 256
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv4_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv4_1/dw"
  top: "conv4_1/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu4_1/dw"
  type: "ReLU"
  bottom: "conv4_1/dw"
  top: "conv4_1/dw"
}
layer {
  name: "conv4_1/sep"
  type: "Convolution"
  bottom: "conv4_1/dw"
  top: "conv4_1/sep"
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv4_1/sep/bn"
  type: "BatchNorm"
  bottom: "conv4_1/sep"
  top: "conv4_1/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu4_1/sep"
  type: "ReLU"
  bottom: "conv4_1/sep"
  top: "conv4_1/sep"
}
layer {
  name: "conv4_2/dw"
  type: "Convolution"
  bottom: "conv4_1/sep"
  top: "conv4_2/dw"
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 256
    stride: 2
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv4_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv4_2/dw"
  top: "conv4_2/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu4_2/dw"
  type: "ReLU"
  bottom: "conv4_2/dw"
  top: "conv4_2/dw"
}
layer {
  name: "conv4_2/sep"
  type: "Convolution"
  bottom: "conv4_2/dw"
  top: "conv4_2/sep"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv4_2/sep/bn"
  type: "BatchNorm"
  bottom: "conv4_2/sep"
  top: "conv4_2/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu4_2/sep"
  type: "ReLU"
  bottom: "conv4_2/sep"
  top: "conv4_2/sep"
}
layer {
  name: "conv5_1/dw"
  type: "Convolution"
  bottom: "conv4_2/sep"
  top: "conv5_1/dw"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 512
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_1/dw"
  top: "conv5_1/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_1/dw"
  type: "ReLU"
  bottom: "conv5_1/dw"
  top: "conv5_1/dw"
}
layer {
  name: "conv5_1/sep"
  type: "Convolution"
  bottom: "conv5_1/dw"
  top: "conv5_1/sep"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_1/sep/bn"
  type: "BatchNorm"
  bottom: "conv5_1/sep"
  top: "conv5_1/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_1/sep"
  type: "ReLU"
  bottom: "conv5_1/sep"
  top: "conv5_1/sep"
}
layer {
  name: "conv5_2/dw"
  type: "Convolution"
  bottom: "conv5_1/sep"
  top: "conv5_2/dw"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 512
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_2/dw"
  top: "conv5_2/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_2/dw"
  type: "ReLU"
  bottom: "conv5_2/dw"
  top: "conv5_2/dw"
}
layer {
  name: "conv5_2/sep"
  type: "Convolution"
  bottom: "conv5_2/dw"
  top: "conv5_2/sep"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_2/sep/bn"
  type: "BatchNorm"
  bottom: "conv5_2/sep"
  top: "conv5_2/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_2/sep"
  type: "ReLU"
  bottom: "conv5_2/sep"
  top: "conv5_2/sep"
}
layer {
  name: "conv5_3/dw"
  type: "Convolution"
  bottom: "conv5_2/sep"
  top: "conv5_3/dw"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 512
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_3/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_3/dw"
  top: "conv5_3/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_3/dw"
  type: "ReLU"
  bottom: "conv5_3/dw"
  top: "conv5_3/dw"
}
layer {
  name: "conv5_3/sep"
  type: "Convolution"
  bottom: "conv5_3/dw"
  top: "conv5_3/sep"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_3/sep/bn"
  type: "BatchNorm"
  bottom: "conv5_3/sep"
  top: "conv5_3/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_3/sep"
  type: "ReLU"
  bottom: "conv5_3/sep"
  top: "conv5_3/sep"
}
layer {
  name: "conv5_4/dw"
  type: "Convolution"
  bottom: "conv5_3/sep"
  top: "conv5_4/dw"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 512
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_4/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_4/dw"
  top: "conv5_4/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_4/dw"
  type: "ReLU"
  bottom: "conv5_4/dw"
  top: "conv5_4/dw"
}
layer {
  name: "conv5_4/sep"
  type: "Convolution"
  bottom: "conv5_4/dw"
  top: "conv5_4/sep"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_4/sep/bn"
  type: "BatchNorm"
  bottom: "conv5_4/sep"
  top: "conv5_4/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_4/sep"
  type: "ReLU"
  bottom: "conv5_4/sep"
  top: "conv5_4/sep"
}
layer {
  name: "conv5_5/dw"
  type: "Convolution"
  bottom: "conv5_4/sep"
  top: "conv5_5/dw"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 512
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_5/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_5/dw"
  top: "conv5_5/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_5/dw"
  type: "ReLU"
  bottom: "conv5_5/dw"
  top: "conv5_5/dw"
}
layer {
  name: "conv5_5/sep"
  type: "Convolution"
  bottom: "conv5_5/dw"
  top: "conv5_5/sep"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_5/sep/bn"
  type: "BatchNorm"
  bottom: "conv5_5/sep"
  top: "conv5_5/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_5/sep"
  type: "ReLU"
  bottom: "conv5_5/sep"
  top: "conv5_5/sep"
}
layer {
  name: "conv5_6/dw"
  type: "Convolution"
  bottom: "conv5_5/sep"
  top: "conv5_6/dw"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 512
    stride: 2
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_6/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_6/dw"
  top: "conv5_6/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_6/dw"
  type: "ReLU"
  bottom: "conv5_6/dw"
  top: "conv5_6/dw"
}
layer {
  name: "conv5_6/sep"
  type: "Convolution"
  bottom: "conv5_6/dw"
  top: "conv5_6/sep"
  convolution_param {
    num_output: 1024
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv5_6/sep/bn"
  type: "BatchNorm"
  bottom: "conv5_6/sep"
  top: "conv5_6/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu5_6/sep"
  type: "ReLU"
  bottom: "conv5_6/sep"
  top: "conv5_6/sep"
}
layer {
  name: "conv6/dw"
  type: "Convolution"
  bottom: "conv5_6/sep"
  top: "conv6/dw"
  convolution_param {
    num_output: 1024
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 1024
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv6/dw/bn"
  type: "BatchNorm"
  bottom: "conv6/dw"
  top: "conv6/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu6/dw"
  type: "ReLU"
  bottom: "conv6/dw"
  top: "conv6/dw"
}
layer {
  name: "conv6/sep"
  type: "Convolution"
  bottom: "conv6/dw"
  top: "conv6/sep"
  convolution_param {
    num_output: 1024
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "conv6/sep/bn"
  type: "BatchNorm"
  bottom: "conv6/sep"
  top: "conv6/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu6/sep"
  type: "ReLU"
  bottom: "conv6/sep"
  top: "conv6/sep"
}
layer {
  name: "relu6/sep/conv_down"
  type: "Convolution"
  bottom: "conv6/sep"
  top: "relu6/sep/conv_down"
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "relu6/sep/conv_down/bn"
  type: "BatchNorm"
  bottom: "relu6/sep/conv_down"
  top: "relu6/sep/conv_down"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu6/sep/relu_down"
  type: "ReLU"
  bottom: "relu6/sep/conv_down"
  top: "relu6/sep/conv_down"
}
layer {
  name: "relu6/sep/relu_down/up2"
  type: "Deconvolution"
  bottom: "relu6/sep/conv_down"
  top: "relu6/sep/relu_down/up2"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 4
    group: 256
    stride: 2
    weight_filler {
      type: "bilinear"
    }
  }
}
layer {
  name: "relu6/sep/relu_down/up2/up4"
  type: "Deconvolution"
  bottom: "relu6/sep/relu_down/up2"
  top: "relu6/sep/relu_down/up2/up4"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 4
    group: 256
    stride: 2
    weight_filler {
      type: "bilinear"
    }
  }
}
layer {
  name: "relu6/sep/relu_down/up2/up4/up8"
  type: "Deconvolution"
  bottom: "relu6/sep/relu_down/up2/up4"
  top: "relu6/sep/relu_down/up2/up4/up8"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 4
    group: 256
    stride: 2
    weight_filler {
      type: "bilinear"
    }
  }
}
layer {
  name: "relu3_1/sep/conv_shortcut"
  type: "Convolution"
  bottom: "conv3_1/sep"
  top: "relu3_1/sep/conv_shortcut"
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "relu3_1/sep/conv_shortcut/bn"
  type: "BatchNorm"
  bottom: "relu3_1/sep/conv_shortcut"
  top: "relu3_1/sep/conv_shortcut"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "relu3_1/sep/relu_shortcut"
  type: "ReLU"
  bottom: "relu3_1/sep/conv_shortcut"
  top: "relu3_1/sep/conv_shortcut"
}
layer {
  name: "cat_block"
  type: "Concat"
  bottom: "relu6/sep/relu_down/up2/up4/up8"
  bottom: "relu3_1/sep/conv_shortcut"
  top: "cat_block"
}
layer {
  name: "ctx_block1/dw"
  type: "Convolution"
  bottom: "cat_block"
  top: "ctx_block1/dw"
  convolution_param {
    num_output: 320
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 320
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "ctx_block1/dw/bn"
  type: "BatchNorm"
  bottom: "ctx_block1/dw"
  top: "ctx_block1/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "ctx_block1/dw/relu"
  type: "ReLU"
  bottom: "ctx_block1/dw"
  top: "ctx_block1/dw"
}
layer {
  name: "ctx_block1/sep"
  type: "Convolution"
  bottom: "ctx_block1/dw"
  top: "ctx_block1/sep"
  convolution_param {
    num_output: 320
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "ctx_block1/sep/bn"
  type: "BatchNorm"
  bottom: "ctx_block1/sep"
  top: "ctx_block1/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "ctx_block1/sep/relu"
  type: "ReLU"
  bottom: "ctx_block1/sep"
  top: "ctx_block1/sep"
}
layer {
  name: "ctx_block2/dw"
  type: "Convolution"
  bottom: "ctx_block1/sep"
  top: "ctx_block2/dw"
  convolution_param {
    num_output: 320
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 320
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "ctx_block2/dw/bn"
  type: "BatchNorm"
  bottom: "ctx_block2/dw"
  top: "ctx_block2/dw"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "ctx_block2/dw/relu"
  type: "ReLU"
  bottom: "ctx_block2/dw"
  top: "ctx_block2/dw"
}
layer {
  name: "ctx_block2/sep"
  type: "Convolution"
  bottom: "ctx_block2/dw"
  top: "ctx_block2/sep"
  convolution_param {
    num_output: 320
    bias_term: false
    pad: 0
    kernel_size: 1
    group: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
    dilation: 1
  }
}
layer {
  name: "ctx_block2/sep/bn"
  type: "BatchNorm"
  bottom: "ctx_block2/sep"
  top: "ctx_block2/sep"
  batch_norm_param {
    scale_bias: true
  }
}
layer {
  name: "ctx_block2/sep/relu"
  type: "ReLU"
  bottom: "ctx_block2/sep"
  top: "ctx_block2/sep"
}
layer {
  name: "ctx_final"
  type: "Convolution"
  bottom: "ctx_block2/sep"
  top: "ctx_final"
  convolution_param {
    num_output: 19
    pad: 0
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "ctx_final/up16"
  type: "Deconvolution"
  bottom: "ctx_final"
  top: "ctx_final/up16"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 19
    bias_term: false
    pad: 1
    kernel_size: 4
    group: 19
    stride: 2
    weight_filler {
      type: "bilinear"
    }
  }
}
layer {
  name: "ctx_output"
  type: "Deconvolution"
  bottom: "ctx_final/up16"
  top: "ctx_output"
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 19
    bias_term: false
    pad: 1
    kernel_size: 4
    group: 19
    stride: 2
    weight_filler {
      type: "bilinear"
    }
  }
}
layer {
  name: "loss"
  type: "SoftmaxWithLoss"
  bottom: "ctx_output"
  bottom: "label"
  top: "loss"
  propagate_down: true
  propagate_down: false
  loss_param {
    ignore_label: 255
    normalization: VALID
  }
}
layer {
  name: "accuracy/top1"
  type: "Accuracy"
  bottom: "ctx_output"
  bottom: "label"
  top: "accuracy/top1"
  include {
    phase: TEST
  }
  accuracy_param {
    ignore_label: 255
  }
}
layer {
  name: "accuracy/top5"
  type: "Accuracy"
  bottom: "ctx_output"
  bottom: "label"
  top: "accuracy/top5"
  include {
    phase: TEST
  }
  accuracy_param {
    top_k: 5
    ignore_label: 255
  }
}

This is my import configuration file: "tidl_import_mobilesegnett1-1.0_mod.txt"

modelType = 0
numParamBits = 16
numFeatureBits = 16
inputNetFile = "../../test/testvecs/models/public/caffe/mod/mobilesegnett1-1.0/deploy.prototxt"
inputParamsFile = "../../test/testvecs/models/public/caffe/mod/mobilesegnett1-1.0/multiod_mobilesegnett1-1.0_iter_420000.caffemodel"
outputNetFile = "../../test/testvecs/config/tidl_models/caffe/tidl_net_mod_mobilesegnett1-1.0_320x320.bin"
outputParamsFile = "../../test/testvecs/config/tidl_models/caffe/tidl_io_mod_mobilesegnett1-1.0_320x320_"
inDataFormat = 0
perfSimConfig = ../../test/testvecs/config/import/device_config.cfg
inData = "../../test/testvecs/config/segmentation_list_mod.txt"
postProcType = 3

This is my inference configuration file: "tidl_infer_mobilesegnett1-1.0_mod.txt"

inFileFormat= 2
postProcType = 3
netBinFile = "testvecs/config/tidl_models/caffe/tidl_net_mod_mobilesegnett1-1.0_320x320.bin"
ioConfigFile = "testvecs/config/tidl_models/caffe/tidl_io_mod_mobilesegnett1-1.0_320x320_1.bin"
outData = "testvecs/output/mod_mobilesegnett1-1.0_320x320.bin"
inData = "testvecs/config/segmentation_list_mod.txt"
numFrames = 1
totNumClasses = 20
debugTraceLevel = 0
writeTraceLevel = 0

I just updated to version( tidl_j7_08_01_00_05) , failed to use "pc_dSP_test_dl_algo.out" for imported model inference:

Error at line: 741 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_08_01_00_05/ti_dl/test/src/pc_linux/../tidl_tb_utils.c, of function : tidl_tb_dataConvert
Error at line: 741 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_08_01_00_05/ti_dl/test/src/pc_linux/../tidl_tb_utils.c, of function : tidl_tb_dataConvert
Error at line: 741 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_08_01_00_05/ti_dl/test/src/pc_linux/../tidl_tb_utils.c, of function : tidl_tb_dataConvert
Error at line: 741 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_08_01_00_05/ti_dl/test/src/pc_linux/../tidl_tb_utils.c, of function : tidl_tb_dataConvert
Error at line: 741 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_08_01_00_05/ti_dl/test/src/pc_linux/../tidl_tb_utils.c, of function : tidl_tb_dataConvert

Regards,

Damon

0 kumar.desappan over 4 years ago in reply to Damon@INVO

TI__Mastermind 22145 points

By "Device" I mean the TDA4V (EVM).

For a new version you need to import the model again, did you import the model with the new version of the tools again?

0 Damon@INVO over 4 years ago in reply to kumar.desappan

Intellectual 750 points

Hi kumar:

This error occurred when I re-imported the model on the new version 8.01.

My recently retrained model, imported and inferred on version 7.3. Inference test error:

Error at line: 483 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_02_00_00_07/ti_dl/test/src/pc_linux/../tidl_rt.c, of function : tidl_copyRTTensorToInBufs
Error at line: 483 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_02_00_00_07/ti_dl/test/src/pc_linux/../tidl_rt.c, of function : tidl_copyRTTensorToInBufs
Error at line: 483 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_02_00_00_07/ti_dl/test/src/pc_linux/../tidl_rt.c, of function : tidl_copyRTTensorToInBufs
Error at line: 483 : in file /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_j7_02_00_00_07/ti_dl/test/src/pc_linux/../tidl_rt.c, of function : tidl_copyRTTensorToInBufs

Import and inferred on version 8.01. Inference test error:

None of the above model imports reported any errors.

Regards,

Damon

0 Damon@INVO over 4 years ago in reply to kumar.desappan

Intellectual 750 points

It's not my job to import models to run on TDA4(EVM), so I don't know much about it.

If neither the imported model nor the inference tests report any errors, will the imported model work in TDA4?

0 kumar.desappan over 4 years ago in reply to Damon@INVO

TI__Mastermind 22145 points

Could you please share the complete steps and collaterals (Model, config files) to re-produce this issue?

BTW, We have Python API-based import support with open source runtimes Like TFlite / ONNX runtime. Would consider this for import and validation?

github.com/.../edgeai-tidl-tools

Processors

Processors forum

TDA4VM: Importing models and running imported models on target