Keras训练CNN时出现“输入数据耗尽”警告及部分轮次损失/准确率为0的问题求助

大家好，我最近在用Keras训练CNN模型时碰到了一个棘手的问题，想请各位帮忙分析一下。

我用ImageDataGenerator处理数据集，训练集一共有1500张图片，设置的batch size是32，所以计算出来的steps_per_epoch是⌈1500/32⌉=47（最后一个批次是不完整的），通过flow_from_directory()方法加载图片。但训练开始后，部分epoch直接显示损失和准确率为0，还弹出了“输入数据耗尽”的警告，训练也被中断，具体警告如下：

UserWarning: Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least steps_per_epoch * epochs batches. You may need to use the .repeat() function when building your dataset.
self.gen.throw(typ, value, traceback)

以下是我的完整代码：

train_datagen = ImageDataGenerator(rescale=1./255)
valid_datagen = ImageDataGenerator(rescale=1./255)

# 从目录导入数据并转换为批次
train_data = train_datagen.flow_from_directory(train_path,
                                           batch_size=32,
                                           target_size=(224, 224),
                                           class_mode="binary")

valid_data = valid_datagen.flow_from_directory(test_path,
                                           batch_size=32,
                                           target_size=(224, 224),
                                           class_mode="binary")

tf.random.set_seed(42)

model_1 = tf.keras.models.Sequential([
   Input(shape=(224, 224, 3)),
   Conv2D(filters=10,
     kernel_size=3,
     strides=1,
     padding='valid',
     activation='relu'),
   Conv2D(10, 3, activation='relu'),
   Conv2D(10, 3, activation='relu'),
   Flatten(),
   Dense(1, activation='sigmoid')
])

model_1.compile(loss="binary_crossentropy",
            optimizer=tf.keras.optimizers.Adam(),
            metrics=["accuracy"])

history_1 = model_1.fit(train_data,
                    epochs=5,
                    steps_per_epoch=len(train_data),
                    validation_data=valid_data,
                    validation_steps=len(valid_data))

训练的输出日志如下：

Epoch 1/5
/usr/local/lib/python3.11/dist-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.
  self._warn_if_super_not_called()

47/47 ━━━━━━━━━━━━━━━━━━━━ 767s 14s/step - accuracy: 0.6488 - loss: 0.6197 - val_accuracy: 0.7620 - val_loss: 0.4793

Epoch 2/5
47/47 ━━━━━━━━━━━━━━━━━━━━ 0s 426us/step - accuracy: 0.0000e+00 - loss: 0.0000e+00

Epoch 3/5
/usr/lib/python3.11/contextlib.py:158: UserWarning: Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches. You may need to use the `.repeat()` function when building your dataset.
  self.gen.throw(typ, value, traceback)

47/47 ━━━━━━━━━━━━━━━━━━━━ 48s 224ms/step - accuracy: 0.8120 - loss: 0.4131 - val_accuracy: 0.8600 - val_loss: 0.3441

Epoch 4/5
47/47 ━━━━━━━━━━━━━━━━━━━━ 3s 54ms/step - accuracy: 0.0000e+00 - loss: 0.0000e+00

Epoch 5/5
47/47 ━━━━━━━━━━━━━━━━━━━━ 11s 219ms/step - accuracy: 0.8336 - loss: 0.3814 - val_accuracy: 0.8880 - val_loss: 0.3042

有没有大佬能帮我看看这是什么原因导致的？该怎么解决这个问题呢？

备注：内容来源于stack exchange，提问作者Hamza Azhar