[onnx-tools] Show tensor shape in onnx-dump

[seanshpark]

onnx-dump.py doesn't show tensor info (shape, dtype) let's revise this tool to show them.

[seanshpark]

when dump model.graph, there is value_info like

value_info {
  name: "/Add_1_output_0"
  type {
    tensor_type {
      elem_type: 1
      shape {
        dim {
          dim_value: 1
        }
        dim {
          dim_value: 2
        }
        dim {
          dim_value: 3
        }
        dim {
          dim_value: 3
        }
      }
    }
  }
}

[shs-park]

when dump model.graph, there is value_info like

• shape: dim_values
• dtype: elem_type

Is it correct..?

[seanshpark]

Is it correct..?

Yup :)

shape: [1,2,3,5] (rank 4) dtype is elem_type = 1

• from googling, https://github.com/onnx/onnx/blob/001550a325bc5ca0d3d0b09395d820439cb3bb37/onnx/onnx.in.proto#L501C1-L514C25

  message TensorProto {
  enum DataType {
  UNDEFINED = 0;
  // Basic types.
  FLOAT = 1;   // float
  UINT8 = 2;   // uint8_t
  INT8 = 3;    // int8_t
  UINT16 = 4;  // uint16_t
  INT16 = 5;   // int16_t
  INT32 = 6;   // int32_t
  INT64 = 7;   // int64_t
  STRING = 8;  // string
  BOOL = 9;    // bool

Type is defined like

message TypeProto {

  message Tensor {
    // This field MUST NOT have the value of UNDEFINED
    // This field MUST have a valid TensorProto.DataType value
    // This field MUST be present for this version of the IR.
    optional int32 elem_type = 1;
    optional TensorShapeProto shape = 2;
  }

[icodo98]

shape: [1,2,3,5] (rank 4) dtype is elem_type = 1

May I ask whether we should consider dynamic tensor shapes in this issue?

IMHO, dim_param in tensor.dim contains this information:

• https://github.com/onnx/onnx/blob/001550a325bc5ca0d3d0b09395d820439cb3bb37/onnx/onnx.in.proto#L693-L706

[seanshpark]

May I ask whether we should consider dynamic tensor shapes in this issue?

Maybe yes, but not now. This issue is to solve my internal problem with ONNX model.

[seanshpark]

done