tf2.0 ValueError:You are trying to load a weight file containing 2 layers into a model with 0 layers

tf2.0 ValueError:You are trying to load a weight file containing 2 layers into a model with 0 layers

tensorflow2.0版本，通过继承 tf.keras.Model 类自定义网络结构，当加载之前训练过的模型参数时报错 ValueError:You are trying to load a weight file containing 2 layers into a model with 0 layers。

网上有说改成
model.load_weights('model_name.h5',by_name=True)
试了下还是报错。

后来在github的一些demo里发现，在加载权重前将模型运行一遍将模型初始化就可以了。

model = MyModel()  # 实例化
s(shape=INPUT_SHAPE))  # 随便用个输入跑一下，初始化模型
model.load_weights('model_name.h5')  # 加载权重

最后吐糟一下 tensorflow2.0 坑有点多