tensorflow - Tensorflow中的标签形状不匹配

Question

我是张量流的初学者。我想建立一个简单的模型，但我得到了这个错误。我认为这是因为标签，但我不知道如何修复它。我使用 tf.data.Dataset 从目录文件中构建我的数据集。

这是数据集：

访问：https ://i.stack.imgur.com/H2EQT.jpg

数据

• --------class1：[x.jpeg ...]
• --------class2：[y.png ....]
• .
• .
• .
• --------class10：[z.jpg ...]

    data_dir = os.path.join(os.path.dirname('D:/Downloads/Image data set/'), 'raw-img')
    data_dir = pathlib.Path(data_dir)
    image_count = len(list(data_dir.glob('*/*.*')))

    list_ds = tf.data.Dataset.list_files(str(data_dir / '*/*'), shuffle=False)
    list_ds = list_ds.shuffle(image_count, reshuffle_each_iteration=False)

    train_size = int(image_count * 0.8)
    test_size = int(image_count * 0.1)
    val_size = int(image_count * 0.1)
    
    train_ds = list_ds.take(train_size)
    val_ds = list_ds.skip(train_size)
    test_ds = val_ds.skip(test_size)
    val_ds = val_ds.take(test_size)

    def parse_image(filename):
      parts = tf.strings.split(filename, os.sep)
      label = tf.cast(parts[-2] == class_names, tf.float32)
      label = tf.argmax(label)
    
      image = tf.io.read_file(filename)
      image = tf.image.decode_jpeg(image, channels=3)
      image = tf.image.convert_image_dtype(image, tf.float32)
      image = tf.reshape(image, [32, 150, 150, 3])
      return image, labels
    
    AUTOTUNE = tf.data.experimental.AUTOTUNE
    
    train_ds = train_ds.map(parse_image, num_parallel_calls=AUTOTUNE)
    val_ds = val_ds.map(parse_image, num_parallel_calls=AUTOTUNE)
    test_ds = test_ds.map(parse_image, num_parallel_calls=AUTOTUNE)
    
    model = tf.keras.Sequential([
        kr.layers.Conv2D(16, 3, activation='relu', input_shape=(150, 150 ,3)),
        kr.layers.MaxPooling2D(),
        kr.layers.Conv2D(32, 3, activation='relu'),
        kr.layers.MaxPooling2D(),
        kr.layers.Conv2D(64, 3, activation='relu'),
        kr.layers.MaxPooling2D(),
        kr.layers.Flatten(),
        kr.layers.Dense(128, activation='softmax'),
        kr.layers.Dense(10)
    ])
    
    model.compile(
      optimizer='adam',
      loss='sparse_categorical_crossentropy',
      metrics=['accuracy'])
    
    model.fit(
      train_ds,
      epochs=3
    )

错误：这是我想拟合模型时的错误。

Train for 20943 steps
Epoch 1/3
    1/20943 [..............................] - ETA: 1:14:41
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-38-c408313d0649> in <module>
      1 model.fit(
      2   train_ds,
----> 3   epochs=3
      4 )

    
.
.
.
.

ValueError: Shape mismatch: The shape of labels (received (320,)) should equal the shape of logits except for the last dimension (received (32, 10)).

标签： tensorflowmachine-learningkeraslabelreshape

Answer 1

通过使用批量大小和 step_per_epoch 变量解决了这个问题。

train_ds = train_ds.map(parse_image, num_parallel_calls=AUTOTUNE)
train_ds = train_ds.cache()
train_ds = train_ds.batch(BATCH_SIZE, drop_remainder=True).repeat()
train_ds = train_ds.prefetch(buffer_size=AUTOTUNE)

和训练

history = model.fit(train_ds, steps_per_epoch = train_size // BATCH_SIZE,
        epochs = EPOCHS, batch_size = BATCH_SIZE,
        validation_data = val_ds, validation_steps = val_size // BATCH_SIZE)

并解决了。