TensorFlow 2冻结模型注意事项

今天在尝试冻结一个TensorFlow模型时遇到错误：

ValueError: Input 0 of node embedding_model/resnet50v2/conv2_block1_preact_bn/AssignNewValue/AssignVariableOp was passed float from embedding_model/resnet50v2/conv2_block1_preact_bn/FusedBatchNormV3/ReadVariableOp/resource:0 incompatible with expected resource.

冻结模型的代码来自这里：

Save, Load and Inference From TensorFlow 2.x Frozen Graph

Hello Underworld.

Lei Mao's Log BookTwitter

模型本身是基于TensorFlow的 keras.app，所以模型出问题的可能性很小。多番尝试之后终于找到了问题所在。