onnx模型解析

image.png

ir_version: 3
producer_name: "onnx.utils.extract_model"
graph {
  node {
    input: "Input3"
    input: "Constant339"
    output: "Minus340_Output_0"
    name: "Minus340"
    op_type: "Sub"
    doc_string: ""
    domain: ""
  }
  node {
    input: "Minus340_Output_0"
    input: "Constant343"
    output: "Block352_Output_0"
    name: "Block352"
    op_type: "Div"
    doc_string: ""
    domain: ""
  }
  name: "Extracted from {CNTKGraph}"
  initializer {
    data_type: 1
    float_data: 127.5
    name: "Constant339"
  }
  initializer {
    data_type: 1
    float_data: 255.0
    name: "Constant343"
  }
  input {
    name: "Input3"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  output {
    name: "Block352_Output_0"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  value_info {
    name: "Minus340_Output_0"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
  value_info {
    name: "Block352_Output_0"
    type {
      tensor_type {
        elem_type: 1
        shape {
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 1
          }
          dim {
            dim_value: 64
          }
          dim {
            dim_value: 64
          }
        }
      }
    }
  }
}
opset_import {
  domain: ""
  version: 7
}

# 了解了onnx的结构后，我们可以根据它的结构将其拆分成多个单节点的onnx模型，以便于对整体模型的单个节点进行测试和分析。
import onnx
from onnx import helper,numpy_helper




#获取对应输入信息
def getInputTensorValueInfo(input_name,model):
    in_tvi = []
    for name in input_name:
        for params_input in model.graph.input:
            if params_input.name == name:
               in_tvi.append(params_input)
        for inner_output in model.graph.value_info:
            if inner_output.name == name:
                in_tvi.append(inner_output)
    return in_tvi

#获取对应输出信息
def getOutputTensorValueInfo(output_name,model):
    out_tvi = []
    for name in output_name:
        out_tvi = [inner_output for inner_output in model.graph.value_info if inner_output.name == name]
        if name == model.graph.output[0].name:
            out_tvi.append(model.graph.output[0])
    return out_tvi

#获取对应超参数值
def getInitTensorValue(input_name,model):
    init_t = []
    for name in input_name:
        init_t = [init for init in model.graph.initializer if init.name == name]
    return init_t

#构建单个节点onnx模型
def createSingelOnnxModel(ModelPath,nodename,SaveType="",SavePath=""):
    model = loadOnnxModel(str(ModelPath))
    Node,input_name,output_name = getNodeAndIOname(nodename,model)
    in_tvi = getInputTensorValueInfo(input_name,model)
    out_tvi = getOutputTensorValueInfo(output_name,model)
    init_t = getInitTensorValue(input_name,model)

    graph_def = helper.make_graph(
                [Node],
                nodename,
                inputs=in_tvi,  # 输入
                outputs=out_tvi,  # 输出
                initializer=init_t,  # initalizer
            )
    model_def = helper.make_model(graph_def, producer_name='onnx-example')
    print(nodename+"onnx模型生成成功！")

# 获取整个ONNX模型的一些信息





#获取节点名列表
def getNodeNameList(model):
    NodeNameList = []
    for i in range(len(model.graph.node)):
        NodeNameList.append(model.graph.node[i].name)
    return NodeNameList





def get_node_attributes(node):
    return node.attribute


def get_node_inputs(node):
    return node.input


def get_node_outputs(node):
    return node.output


def show_weight(weight):
    print("="*10, "details of weight: ", weight.name, "="*10)
    print("data type: ", weight.data_type)
    print("shape: ", weight.dims)
    data_numpy = numpy_helper.to_array(weight)
    # data_numpy = np.frombuffer(weight.raw_data, dtype=xxx)
    # print("detail data:", data_numpy)
    print("="*40)



# onnx.utils.extract_model("emotion-ferplus-7.onnx","mini_model.onnx",["Input3"],["Block352_Output_0"])

model = onnx.load("mini_model.onnx")
# print(model.ir_version)   # IR的版本
# print(model.producer_name)   #
# print(model.opset_import)   # opset 版本信息


# graph   
# graph中有node(NodeProto类型)，input(ValueInfoProto类型)，output(ValueInfoProto类型)和initializer(TensorProto类型)
# 其中node中存放着模型中的所有计算节点，input中存放着模型所有的输入节点，output存放着模型所有的输出节点，initializer存放着模型所有的权重；
#value_info存放了各个tensor 的信息
# node 通过input和output的指向关系，描绘出一个深度学习模型的拓扑图
# for node in model.graph.node:
#     print(node)

#获取节点和节点的输入输出名列表，一般节点的输入将来自于上一层的输出放在列表前面，参数放在列表后面
def getNode(nodename,model):
    for i in range(len(model.graph.node)):
        if model.graph.node[i].name == nodename:
            Node = model.graph.node[i]
            input_name = model.graph.node[i].input
            output_name = model.graph.node[i].output
    return Node,input_name,output_name


    
for node in model.graph.node:
    print(get_node_inputs(node))
    print(get_node_outputs(node))
    print(get_node_attributes(node))


# print(model.graph.input)
# print(model.graph.output)


# for value_inifo in  model.graph.value_info:
#     print(value_inifo)


# for initializer in  model.graph.initializer:
#     print(initializer)



#获取节点数量
print(len(model.graph.node))




# with open("model.txt","w") as f:
#     f.write(str(model))
    
# print(model.graph)

print(getNodeNameList(model))

https://github.com/ZhangGe6/onnx-modifier/tree/master
https://github.com/bindog/onnx-surgery/blob/master/surgery.py
https://bindog.github.io/blog/2020/03/13/deep-learning-model-convert-and-depoly/
https://www.zhihu.com/question/386526462
https://blog.csdn.net/ChuiGeDaQiQiu/article/details/123794387