将 2D 图像馈送到 TensorFlow CNN 以进行图像分类

Question

感谢您提供帮助。我正在尝试使用 TensorFlow 训练卷积神经网络来预测我的系统的二进制方面。我有大约 1000 个 ndarray，大小为 400x400x3，持有浮动。这类似于（我认为）RGB 图像；我的图像的三个方面不是颜色，而是三个不同功能的输出，但这在这里不重要。这些 ndarray 中的每一个都与一个二进制标签相关联——0 或 1。我是 TensorFlow 的新手，虽然在机器学习原理上相当扎实，所以非常感谢有关解码错误的指导。我认为我排列图层的方式很奇怪，但我尝试非常密切地按照教程进行操作，所以我不知道为什么。

我的代码如下：

no_of_samples = 1035
train_batches = 30
BATCH_SIZE = 23
SHUFFLE_BUFFER_SIZE = 50

data, labels = np.load("total_image_data.npy", allow_pickle=True), np.load("labels.npy", allow_pickle=True)

#divide into test and train sets

train_indices = np.random.choice([i for i in range(no_of_samples)], size=train_batches*BATCH_SIZE)
test_indices = [i for i in range(no_of_samples) if i not in train_indices]

data_train, labels_train = [data[i] for i in train_indices], [labels[i] for i in train_indices]
data_test, labels_test = [data[i] for i in test_indices], [labels[i] for i in test_indices]

train_dataset = tf.data.Dataset.from_tensor_slices((data_train, labels_train))
test_dataset = tf.data.Dataset.from_tensor_slices((data_test, labels_test))

train_dataset = train_dataset.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
test_dataset = test_dataset.batch(BATCH_SIZE)



model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(200, 5, strides=3, activation='relu', input_shape=(BATCH_SIZE, 400, 400, 3)),
    tf.keras.layers.Conv2D(100, 5, strides=2, activation="relu"),
    tf.keras.layers.Conv2D(50, 5, activation="relu"),
    tf.keras.layers.Conv2D(25, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(3),
    tf.keras.layers.Conv2D(50, 3, activation="relu"),
    tf.keras.layers.Conv2D(25, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(3),
    tf.keras.layers.Conv2D(50, 2, activation="relu"),
    tf.keras.layers.Conv2D(25, 2, activation="relu"),
    
    tf.keras.layers.GlobalMaxPooling2D(),

    # Finally, we add a classification layer.
    tf.keras.layers.Dense(2)
])

model.compile(optimizer=tf.keras.optimizers.RMSprop(),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['sparse_categorical_accuracy'])

model.fit(labels_train, data_train, epochs=10, batch_size=BATCH_SIZE)

model.evaluate(labels_test, data_test)

我运行时遇到的错误如下：

Traceback (most recent call last):
  File "training.py", line 36, in <module>
    model = tf.keras.Sequential([
  File "/local/**/ac3/lib/python3.8/site-packages/tensorflow/python/training/tracking/base.py", line 530, in _method_wrapper
    result = method(self, *args, **kwargs)
  File "/local/**/ac3/lib/python3.8/site-packages/keras/utils/traceback_utils.py", line 67, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/local/**/ac3/lib/python3.8/site-packages/keras/engine/input_spec.py", line 213, in assert_input_compatibility
    raise ValueError(f'Input {input_index} of layer "{layer_name}" '
ValueError: Input 0 of layer "max_pooling2d" is incompatible with the layer: expected ndim=4, found ndim=5. Full shape received: (None, 23, 58, 58, 25)

任何帮助解码此内容将不胜感激。

更新：新的令人兴奋的错误！在一些慷慨的社区成员的帮助下，我的代码现在看起来像这样：

import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

no_of_samples = 1035
BATCH_SIZE = 16
SHUFFLE_BUFFER_SIZE = 50

data, labels = np.load("total_image_data.npy", allow_pickle=True), np.load("labels.npy", allow_pickle=True)

print(np.shape(data)) ##this outputs (1035, 400, 400, 3)
print(np.shape(labels)) ##this outputs (1035, )



#divide into test and train sets


dataset = tf.data.Dataset.from_tensor_slices((data, labels)).shuffle(SHUFFLE_BUFFER_SIZE)
test_dataset = dataset.take(100).batch(BATCH_SIZE)
train_dataset = dataset.skip(100).batch(BATCH_SIZE)

model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(200, 5, strides=3, activation='relu', input_shape=(400, 400, 3)),
    tf.keras.layers.Conv2D(100, 5, strides=2, activation="relu"),
    tf.keras.layers.Conv2D(50, 5, activation="relu"),
    tf.keras.layers.Conv2D(25, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(3),
    tf.keras.layers.Conv2D(50, 3, activation="relu"),
    tf.keras.layers.Conv2D(25, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(3),
    tf.keras.layers.Conv2D(50, 2, activation="relu"),
    tf.keras.layers.Conv2D(25, 2, activation="relu"),
    
    tf.keras.layers.GlobalMaxPooling2D(),

    # Finally, we add a classification layer.
    tf.keras.layers.Dense(1)
])

print('Labels shape -->',labels.shape)
print('Labels -->', labels)

model.compile(optimizer=tf.keras.optimizers.RMSprop(),
              loss=tf.keras.losses.BinaryCrossentropy(),
              metrics=['accuracy'])

model.fit(train_dataset, epochs=10)
model.evaluate(test_dataset)

这编译并运行良好，但报告了 nan 的损失和 58% 的跨时期的不变准确度（我正在寻找的东西的基线流行率为 38%）。再次，我投身于你的怜悯。

Answer 1

我认为您需要更改一些内容：模型的输入形状不需要batch_size，它将在训练期间进行推断。将其更改为(400, 400, 3)。其次，如果您使用二进制标签，则需要将损失函数更改为tf.keras.losses.BinaryCrossentropy，将度量更改为tf.keras.metrics.BinaryAccuracy 或简单的accuracy。此外，您的输出层应该有一个输出节点而不是两个：tf.keras.layers.Dense(1)

这是一个基于您的代码的运行示例：

import numpy as np
import tensorflow as tf

no_of_samples = 250
BATCH_SIZE = 16
SHUFFLE_BUFFER_SIZE = 50

data, labels = np.random.random((no_of_samples, 400, 400, 3)), np.random.randint(2, size=no_of_samples)

dataset = tf.data.Dataset.from_tensor_slices((data, labels)).shuffle(SHUFFLE_BUFFER_SIZE)
test_dataset = dataset.take(50).batch(BATCH_SIZE)
train_dataset = dataset.skip(50).batch(BATCH_SIZE)

model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(200, 5, strides=3, activation='relu', input_shape=(400, 400, 3)),
    tf.keras.layers.Conv2D(100, 5, strides=2, activation="relu"),
    tf.keras.layers.Conv2D(50, 5, activation="relu"),
    tf.keras.layers.Conv2D(25, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(3),
    tf.keras.layers.Conv2D(50, 3, activation="relu"),
    tf.keras.layers.Conv2D(25, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(3),
    tf.keras.layers.Conv2D(50, 2, activation="relu"),
    tf.keras.layers.Conv2D(25, 2, activation="relu"),
    
    tf.keras.layers.GlobalMaxPooling2D(),

    # Finally, we add a classification layer.
    tf.keras.layers.Dense(1)
])

model.compile(optimizer=tf.keras.optimizers.RMSprop(),
              loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
              metrics=['accuracy'])

model.fit(train_dataset, epochs=10, validation_data=test_dataset)

print('Labels shape -->',labels.shape)
print('Labels -->', labels)
Labels shape --> (250,)
Labels --> [1 0 0 0 0 1 0 1 0 1 1 1 0 1 1 0 0 1 0 0 1 0 0 1 0 1 1 0 0 1 1 1 0 1 1 0 0
 0 1 0 1 1 1 0 1 1 1 1 0 0 0 0 0 1 1 1 0 1 1 0 0 0 1 1 0 1 1 0 0 1 1 1 0 0
 1 0 1 1 1 1 1 1 1 1 0 0 1 1 0 1 1 1 1 0 1 1 0 0 0 1 0 1 1 1 0 1 0 1 1 0 1
 1 1 1 1 0 0 0 1 0 0 0 1 1 1 0 1 1 1 0 0 0 1 1 1 0 1 0 1 0 1 1 0 1 0 0 1 0
 1 1 1 1 0 0 1 0 0 0 0 0 0 0 0 1 1 0 1 1 1 0 0 0 0 1 1 0 0 1 1 1 0 0 0 0 1
 0 0 0 0 1 1 0 1 1 1 0 0 0 0 1 0 1 1 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0 1
 0 0 1 1 1 1 1 0 0 0 1 0 0 1 0 1 1 1 1 1 0 1 1 1 0 1 1 0]

Answer 2

当损失达到 nan 时，最简单的解释是你的梯度要么爆炸，要么消失。尝试在优化器上使用梯度裁剪（按值或按范数），这基本上不允许梯度超过或超过某个值。

这也会影响网络的性能，但总比损失 nan 好。

使用您当前的代码，实现将如下所示：

model.compile(optimizer=tf.keras.optimizers.RMSprop(clipvalue=0.5),
              loss=tf.keras.losses.BinaryCrossentropy(),
              metrics=['accuracy'])

该值只是一个示例，如果有效，请尝试稍微调整一下。