Keras报错TypeError: L2.__call__()收到意外参数'dtype'求助

错误原因 #

你把正则化器错误传给了kernel_initializer参数。regularizers.l2(0.01)是权重正则化工具，对应参数是kernel_regularizer；而kernel_initializer需要的是权重初始化器（如'glorot_uniform'、RandomNormal()这类）。

修正代码 #

修改出错的Dense层参数即可，修正后的完整相关代码：

next_layer=GlobalAveragePooling2D()(output)
next_layer=Dense(1024, activation='relu')(next_layer)
next_layer=Dense(512, activation='relu')(next_layer)
# 修正：将kernel_initializer替换为kernel_regularizer
next_layer = Dense(128, activation='relu', kernel_regularizer=regularizers.l2(0.01))(next_layer)
next_layer=Dense(2048, activation='relu')(next_layer)

补充说明 #

如果需要同时设置权重初始化器，可额外添加kernel_initializer参数，示例：

next_layer = Dense(128, 
                   activation='relu', 
                   kernel_regularizer=regularizers.l2(0.01),
                   kernel_initializer='glorot_uniform')(next_layer)

内容的提问来源于stack exchange，提问作者Slma