使用TensorFlow加载Stanford Dogs数据集训练分类器遇形状不匹配错误

尝试使用Stanford Dogs数据集练习图像增强并构建图像分类器，但训练模型时持续出现输入形状不匹配错误，推测问题可能与normalize函数中调整图像尺寸的方式有关。

原始代码 #

import tensorflow as tf
import tensorflow_datasets as tfds
import tensorflow_hub as hub
from tensorflow.keras import layers
tfds.disable_progress_bar()

import os
import matplotlib.pyplot as plt
import matplotlib.pylab as plt
import math
import numpy as np
import logging
logger = tf.get_logger()
logger.setLevel(logging.ERROR)

dataset, metadata = tfds.load('stanford_dogs', as_supervised=True, with_info=True)
train_dataset, test_dataset = dataset['train'], dataset['test']

num_train_examples = metadata.splits['train'].num_examples
num_test_examples = metadata.splits['test'].num_examples
class_names = metadata.features['label'].names

IMG_LEN = 224
N_BREEDS = 120
epochs = 50

image_size = (IMG_LEN, IMG_LEN)
def normalize(image, label):
    normalized_image = tf.image.resize(image, image_size)
    normalized_image /= 255
    return normalized_image, label

batch_size = 32

training_batches = train_dataset.cache().shuffle(num_train_examples//4).batch(batch_size).map(normalize).prefetch(1)
testing_batches = test_dataset.cache().shuffle(num_train_examples//4).batch(batch_size).map(normalize).prefetch(1)

rescale = tf.keras.Sequential([
  layers.Rescaling(1./255, input_shape = (IMG_LEN, IMG_LEN,3))
])

data_augmentation = tf.keras.Sequential([
  layers.RandomFlip("horizontal_and_vertical"),
  layers.RandomRotation(0.2),
  layers.RandomZoom(0.1)
])

model = tf.keras.models.Sequential([
    rescale,
    data_augmentation,
    tf.keras.layers.Conv2D(16, (3,3), activation='relu', input_shape=(IMG_LEN, IMG_LEN, 3)),
    tf.keras.layers.MaxPooling2D(2, 2),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(600, activation='relu'),
    tf.keras.layers.Dense(len(class_names), activation=tf.nn.softmax)
    ])


model.compile(optimizer= "adam",
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=["accuracy"])

from tensorflow.python.distribute.cross_device_ops import validate_destinations
history = model.fit(
    train_dataset,
    epochs = epochs, 
    steps_per_epoch = 1200
)

错误信息 #

Epoch 1/50
---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)
<ipython-input-12-a35416ceec02> in <module>
     47     train_dataset,
     48     epochs = epochs,
---> 49     steps_per_epoch = 1200
     50 )

1 frames
/usr/local/lib/python3.7/dist-packages/keras/engine/training.py in tf__train_function(iterator)
     13                 try:
     14                     do_return = True
---> 15                     retval_ = ag__.converted_call(ag__.ld(step_function), (ag__.ld(self), ag__.ld(iterator)), None, fscope)
     16                 except:
     17                     do_return = False

ValueError: in user code:

    File "/usr/local/lib/python3.7/dist-packages/keras/engine/training.py", line 1051, in train_function  *
        return step_function(self, iterator)
    File "/usr/local/lib/python3.7/dist-packages/keras/engine/training.py", line 1040, in step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    File "/usr/local/lib/python3.7/dist-packages/keras/engine/training.py", line 1030, in run_step  **
        outputs = model.train_step(data)
    File "/usr/local/lib/python3.7/dist-packages/keras/engine/training.py", line 889, in train_step
        y_pred = self(x, training=True)
    File "/usr/local/lib/python3.7/dist-packages/keras/utils/traceback_utils.py", line 67, in error_handler
        raise e.with_traceback(filtered_tb) from None
    File "/usr/local/lib/python3.7/dist-packages/keras/engine/input_spec.py", line 264, in assert_input_compatibility
        raise ValueError(f'Input {input_index} of layer "{layer_name}" is '

    ValueError: Input 0 of layer "sequential_5" is incompatible with the layer: expected shape=(None, 224, 224, 3), found shape=(None, None, 3)

问题分析与修复方案 #

核心问题点 #

1. 训练数据集未处理：model.fit传入了原始的train_dataset，而非你已经做了尺寸调整和批处理的training_batches，导致输入图像尺寸不统一。
2. 重复归一化：normalize函数中已经将图像值除以255，模型里的Rescaling层再次缩放，会导致图像值异常。
3. 冗余的input_shape参数：Sequential模型中前面的层已确定输入形状，Conv2D层的input_shape参数多余且可能引发形状冲突。
4. 损失函数参数错误：最后一层用了softmax激活，SparseCategoricalCrossentropy的from_logits应设为False（可省略）。
5. steps_per_epoch设置不合理：手动指定1200步不符合实际数据量，应使用num_train_examples // batch_size。

修改后的完整代码 #

import tensorflow as tf
import tensorflow_datasets as tfds
from tensorflow.keras import layers
tfds.disable_progress_bar()

import matplotlib.pyplot as plt
import numpy as np
import logging
logger = tf.get_logger()
logger.setLevel(logging.ERROR)

# 加载数据集
dataset, metadata = tfds.load('stanford_dogs', as_supervised=True, with_info=True)
train_dataset, test_dataset = dataset['train'], dataset['test']

num_train_examples = metadata.splits['train'].num_examples
num_test_examples = metadata.splits['test'].num_examples
class_names = metadata.features['label'].names

IMG_LEN = 224
epochs = 50
batch_size = 32

# 数据预处理：调整尺寸+归一化
image_size = (IMG_LEN, IMG_LEN)
def normalize(image, label):
    normalized_image = tf.image.resize(image, image_size)
    normalized_image /= 255.0
    return normalized_image, label

# 构建批处理数据集
training_batches = train_dataset.cache().shuffle(num_train_examples//4).batch(batch_size).map(normalize).prefetch(tf.data.AUTOTUNE)
testing_batches = test_dataset.cache().batch(batch_size).map(normalize).prefetch(tf.data.AUTOTUNE)

# 图像增强层
data_augmentation = tf.keras.Sequential([
  layers.RandomFlip("horizontal"),
  layers.RandomRotation(0.2),
  layers.RandomZoom(0.1)
])

# 构建模型
model = tf.keras.models.Sequential([
    data_augmentation,
    tf.keras.layers.Conv2D(16, (3,3), activation='relu'),
    tf.keras.layers.MaxPooling2D(2, 2),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(600, activation='relu'),
    tf.keras.layers.Dense(len(class_names), activation=tf.nn.softmax)
])

# 编译模型
model.compile(optimizer= "adam",
              loss=tf.keras.losses.SparseCategoricalCrossentropy(),
              metrics=["accuracy"])

# 训练模型
history = model.fit(
    training_batches,
    epochs=epochs,
    steps_per_epoch=num_train_examples // batch_size
)

额外说明 #

• 图像增强中去掉了垂直翻转，因为狗的姿态垂直翻转没有实际意义，可根据需求调整。
• 测试集不需要shuffle，所以去掉了shuffle操作。
• 使用tf.data.AUTOTUNE代替固定的prefetch(1)，让系统自动优化预取数量。

内容的提问来源于stack exchange，提问作者Jackson Roach