嗜甜_513d

```python
g1def = g1.as_graph_def() #g2 g3 g4 同理
with tf.Graph().as_default() as g_combined:
with tf.Session(graph=g_combined) as sess:
x = tf.placeholder(shape,dtype,"placeholder_name")
a = tf.import_graph_def(g1def, input_map={"a_input:0": x}, return_elements=["a_return:0"])

b = tf.import_graph_def(g2def, input_map={"b_input:0": a
, "b_input_2": 'balabala'}, # 如果两个输入
return_elements=["b_return:0"])

c = tf.import_graph_def(g3def, input_map={"c_input:0": b}, return_elements=["c_return:0"])

d = tf.import_graph_def(g4def, input_map={"d_input:0": c}, return_elements=["d_return:0"])

tf.identity(d,"output_combined")

# 定义合并之后的输入输出节点
signature = predict_signature_def(
inputs={'input': g_combined.get_tensor_by_name('placeholder_name:0')},
outputs={'output': g_combined.get_tensor_by_name('output_combined:0')}
)

builder = saved_model_builder.SavedModelBuilder(path)
builder.add_meta_graph_and_variables(
sess=sess,
tags=['TAG'],#随意取 调用时需要
signature_def_map={'serving_default': signature}
, strip_default_attrs=True)
builder.save()
```

tensorflow,存储读取数据结构剖析与合并多个graph，看不懂你掐死我

我又一次开始了“看不懂你掐死我系列”。标题名称是仿照知乎的一篇介绍傅里叶变换的文章起的。当时看完了觉得还真看懂了。可是关上网页再自己想的时候，就有想掐死博主的冲动~~ 为了致...

木木爱吃糖醋鱼

9859 14 10