【onnx】——探索onnx文件

1. 依赖

• protobuf
• onnx模型

2. 探索

1. 采用protobuf的方式。其中 onnx.proto 就是定义的序列化文件
   protoc --decode=onnx.ModelProto onnx.proto < yourfile.onnx

打印出来的结果：

ir_version: 4
producer_name: "pytorch"
producer_version: "1.1"
graph {
  node {
    input: "inputx"
    input: "conv1.weight"
    output: "14"
    op_type: "Conv"
    attribute {
      name: "dilations"
      ints: 1
      ints: 1
      type: INTS
    }
    attribute {
      name: "group"
      i: 1
      type: INT
    }
    attribute {
      name: "kernel_shape"
      ints: 3
      ints: 3
      type: INTS
    }
    attribute {
      name: "pads"
      ints: 1
      ints: 1
      ints: 1
      ints: 1
      type: INTS
    }
    attribute {
      name: "strides"
      ints: 2
      ints: 2
      type: INTS
    }
  }
  node {
    input: "14"
    input: "bn1.weight"
    input: "bn1.bias"
    input: "bn1.running_mean"
    input: "bn1.running_var"
    output: "15"
    op_type: "BatchNormalization"
    attribute {
      name: "epsilon"
      f: 1e-05
      type: FLOAT
    }
    attribute {
      name: "momentum"
      f: 0.9
      type: FLOAT
    }
  }
  node {
    input: "15"
    input: "prelu.weight"
    output: "16"
    op_type: "PRelu"
  }
  node {
    input: "16"
    input: "conv2.weight"
    output: "17"
    op_type: "Conv"
    attribute {
      name: "dilations"
      ints: 1
      ints: 1
      type: INTS
    }
    attribute {
      name: "group"
      i: 1
      type: INT
    }
    attribute {
      name: "kernel_shape"
      ints: 3
      ints: 3
      type: INTS
    }
    attribute {
      name: "pads"
      ints: 1
      ints: 1
      ints: 1
      ints: 1
      type: INTS
    }
    attribute {
      name: "strides"
      ints: 2
      ints: 2
      type: INTS
    }
  }
  node {
    input: "17"
    input: "bn2.weight"
    input: "bn2.bias"
    input: "bn2.running_mean"
    input: "bn2.running_var"
    output: "18"
    op_type: "BatchNormalization"
    attribute {
      name: "epsilon"
      f: 1e-05
      type: FLOAT
    }
    attribute {
      name: "momentum"
      f: 0.9
      type: FLOAT
    }
  }
  node {
    input: "18"
    input: "prelu.weight"
    output: "outputy"
    op_type: "PRelu"
  }
  name: "torch-jit-export"
  initializer {
    dims: 64
    data_type: 1
    name: "bn1.bias"
    raw_data: "\000\000\000\}
input {
    name: "inputx"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 3
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "conv1.weight"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 3
          }
          dim {
            dim_value: 3
          }
          dim {
            dim_value: 3
          }
        }
      }
    }
  }
  input {
    name: "bn1.weight"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "bn1.bias"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "bn1.running_mean"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "bn1.running_var"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "conv2.weight"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 3
          }
          dim {
            dim_value: 3
          }
        }
      }
    }
  }
  input {
    name: "bn2.weight"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "bn2.bias"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "bn2.running_mean"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "bn2.running_var"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  input {
    name: "prelu.weight"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  output {
    name: "outputy"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 16
          }
          dim {
            dim_value: 16
          }
        }
      }
    }
  }
}
opset_import {
  version: 9
}

采用这种方式，推荐找一个只有几层网络的onnx模型，可以清晰看到onnx模型的内部组成，内部结构

2. 采用netron可视化
   这个只需要安装netron之后，就ok了。方便用于检查：数据维度，数值，连接关系等