0

我将模型保存为“已保存模型”。我正在尝试使用freeze graph.py冻结图形， 但它失败并出现错误（dense_1/BiasAdd 不在图形中）所以，我跟踪并打印了图形定义。

有节点def，但图def中没有节点。

node_def {
  name: "dense_1/BiasAdd"
  op: "BiasAdd"
  input: "dense_1/MatMul:product:0"
  input: "dense_1/BiasAdd/ReadVariableOp:value:0"
  attr {
    key: "T"
    value {
      type: DT_FLOAT
    }
  }
  attr {
    key: "_output_shapes"
    value {
      list {
        shape {
          dim {
            size: -1
          }
          dim {
            size: 136
          }
        }
      }
    }
  }
  experimental_debug_info {
    original_node_names: "dense_1/BiasAdd"
  }
}

在 tensorflow 1 中，我可以找到输出节点。

node {
  name: "layer6/logits/BiasAdd"
  op: "BiasAdd"
  input: "layer6/logits/MatMul"
  input: "layer6/logits/bias/read"
  attr {
    key: "T"
    value {
      type: DT_FLOAT
    }
  }
  attr {
    key: "_output_shapes"
    value {
      list {
        shape {
          dim {
            size: -1
          }
          dim {
            size: 136
          }
        }
      }
    }
  }
  attr {
    key: "data_format"
    value {
      s: "NHWC"
    }
  }
}

首先，我想知道为什么tensorflow 2 graph def中没有输出节点。

其次，我想在 tensorflow2 keras 模型中冻结图形。（我可以在 tensorflow 1.15 中做到这一点）

这是模型。

input_tensor = Input(shape=(WIDTH, HEIGHT, DEPTH))
x = Conv2D(32, 3, activation='relu')(input_tensor)
x = BatchNormalization()(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2))(x)
x = Conv2D(64, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = Conv2D(64, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2))(x)
x = Conv2D(64, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = Conv2D(64, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2))(x)
x = Conv2D(128, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = Conv2D(128, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(1, 1))(x)
x = Conv2D(256, 3, activation='relu')(x)
x = BatchNormalization()(x)
x = Flatten()(x)
x = Dense(256, activation='relu')(x)
x = BatchNormalization()(x)
output_tensor = Dense(LANDMARK*2)(x)

model = Model(input_tensor, output_tensor)
4

0 回答 0