Question

#my model
model=Sequential([
    Conv2D(input_shape=(200,360,3),kernel_size=(3,3),filters=64,padding='same'),
    MaxPool2D(pool_size=(2,2)),
    Dropout(rate=0.5),
    Conv2D(kernel_size=(3,3),filters=128,padding='same',activation='relu'),
    Conv2D(kernel_size=(3,3),filters=256,padding='valid',activation='relu'),
    MaxPool2D(pool_size=(2,2)),
    Dropout(rate=0.5),
    Flatten(),
    Dense(units=512,activation='relu'),
    Dropout(rate=0.5),
    Dense(units=256,activation='relu'),
    Dropout(rate=0.5),
    Dense(units=14,activation='softmax')
])

model.add(tf.keras.layers.Reshape((-1,)))

model.compile(optimizer=tf.keras.optimizers.Adam(),
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])

#fit function
model.fit(x_train,y_train,epochs=5)

#Error
InvalidArgumentError:  logits and labels must have the same first dimension, got logits shape [1,14] and labels shape [14]
     [[node sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits (defined at <ipython-input-11-0743e46bb17f>:1) ]] [Op:__inference_train_function_1933]

Function call stack:
train_function

输入图像形状为(400,720,3)。它重塑为(200,360,3)
我将训练数据集放入模型中，编译器说一维的logit和标签形状不相同（logits形状为[1,14]，label形状为[14]）。这些参数是内部变量，所以我无法更改

所以我添加了模型以重塑图层

(model.add(tf.keras.layer.Reshape((-1))))

但这不起作用。

如果您对我的计划有任何疑问，我会向您提供任何信息

keras model.fit（）的错误logit和标签形状不匹配错误

0 个答案: