Keras报错解决：ValueError: Graph disconnected: cannot obtain value for tensor Tensor

---

问题描述：

今天在合并多个Keras model时，遇到错误，记录一下解决方法。

---

抛出错误：

ValueError: Graph disconnected: cannot obtain value for tensor Tensor(“input_14:0”, shape=(None, 24, 9, 1), dtype=float32) at layer “input_14”. The following previous layers were accessed without issue: []

---

原因分析：

查找资料，发现错误原因有三类：

1. 变量名重用：参考这篇文章；
2. 调用模型直接合并错误：参考这个回答；
3. 重建Model时，模型的inputs参数输入不全：参考这个issue；

---

解决方法：

经排查，我遇到的问题属于第3种，解决方法如下：
完整代码：

def mvcnn_model():
'''
down,up,left,right 为 keras model
'''
    n_classes = 2
    data = Input(shape=(24,9,1), name='inps')
    down_out = down.output
    up_out = up.output
    left_out = left.output
    right_out = right.output
    
    down_in = down.input
    up_in = up.input
    left_in = left.input
    right_in = right.input

    merged = concatenate([down_out, up_out, left_out, right_out, data], axis=-1)
    model = Model(inputs=data, outputs=merged)
    return model

出错语句：

  model = Model(inputs=data, outputs=merged)

解决方法：

model = Model(inputs=[down_in, up_in, left_in, right_in, data], outputs=dense4)