使用预测方法时，KERAS CNN模型类型误差

Question

我有一个keras模型，应该采用（150、150、1）灰度图像，因为它的输入并输出一个长度为8。

这是我的模型代码：

from tensorflow.python import keras
model = keras.Sequential([
    keras.layers.Conv2D(filters=32, kernel_size=(3,3), activation="relu", padding='same', input_shape=(150,150,1)),
    keras.layers.MaxPool2D(pool_size=(2,2), padding='same', data_format='channels_last'),
    keras.layers.Conv2D(filters=64, kernel_size=(3,3), activation='relu', padding='same'),
    keras.layers.MaxPool2D(pool_size=(2,2), padding='same', data_format='channels_last'),
    keras.layers.Flatten(),
    keras.layers.Dense(8, activation="softmax")
])

当我尝试使用时.predict（）方法，我得到此错误：

Traceback (most recent call last):
  File "KerasCNN.py", line 152, in <module>
    ga.run()
  File "/home/User/Documents/Projects/2022/Keras_CNN/Trial1/env/lib/python3.6/site-packages/pygad/pygad.py", line 1192, in run
    self.last_generation_fitness = self.cal_pop_fitness()
  File "/home/User/Documents/Projects/2022/Keras_CNN/Trial1/env/lib/python3.6/site-packages/pygad/pygad.py", line 1159, in cal_pop_fitness
    fitness = self.fitness_func(sol, sol_idx)
  File "KerasCNN.py", line 112, in fitness
    prediction = model.predict(g_img)
  File "/home/User/Documents/Projects/2022/Keras_CNN/Trial1/env/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/models.py", line 966, in predict
    return self.model.predict(x, batch_size=batch_size, verbose=verbose)
  File "/home/User/Documents/Projects/2022/Keras_CNN/Trial1/env/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 1813, in predict
    f, ins, batch_size=batch_size, verbose=verbose, steps=steps)
  File "/home/User/Documents/Projects/2022/Keras_CNN/Trial1/env/lib/python3.6/site-packages/tensorflow/python/keras/_impl/keras/engine/training.py", line 1300, in _predict_loop
    index_array = np.arange(num_samples)
TypeError: unsupported operand type(s) for /: 'Dimension' and 'int'

我有一个较早运行的ANN（non-CNN）模型，该模型正常。当我进行一些研究时，我也可以找到有关此错误的任何信息。

这是我用来做出预测的代码：

img = get_image() # (150, 150, 3)
g_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # (150, 150, 1)
g_img = tf.expand_dim(g_img, axis=0)
g_img = tf.expand_dim(g_img, axis=-1) # (1, 150, 150, 1)
prediction = model.predict(g_img)

以下是我的版本号：

tensorflow：1.5.0

python：3.69

numpy：1.19.5

ubuntu：18.04

让我知道我是否可以提供任何其他信息！谢谢！

回答

tf.expand_dim（）用np.expand_dim（）修复了它！

Answer 1

这似乎在TF 1.15上很好地运行：

import cv2
import numpy as np
import tensorflow as tf
from tensorflow.python import keras
print(tf.__version__)
model = keras.Sequential([
    keras.layers.Conv2D(filters=32, kernel_size=(3,3), activation="relu", padding='same', input_shape=(150,150,1)),
    keras.layers.MaxPool2D(pool_size=(2,2), padding='same', data_format='channels_last'),
    keras.layers.Conv2D(filters=64, kernel_size=(3,3), activation='relu', padding='same'),
    keras.layers.MaxPool2D(pool_size=(2,2), padding='same', data_format='channels_last'),
    keras.layers.Flatten(),
    keras.layers.Dense(8, activation="softmax")
])
# Create random image
img = np.zeros([150,150,3], dtype=np.uint8)
img[:,:,0] = np.ones([150,150])*64
img[:,:,1] = np.ones([150,150])*128
img[:,:,2] = np.ones([150,150])*192

g_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
g_img = np.expand_dims(g_img, axis=0)
g_img = np.expand_dims(g_img, axis=-1) # (1, 150, 150, 1)
prediction = model.predict(g_img)
print(prediction.shape)

1.15.2
(1, 8)

Answer 2

在您写的代码中，输入形状为（224,256,1），因此将其更改为（150,150,1）
尝试以下操作：

from tensorflow.python import keras
model = keras.Sequential([
    keras.layers.Conv2D(filters=32, kernel_size=(3,3), activation="relu", padding='same', input_shape=(150,150,1)),
    keras.layers.MaxPool2D(pool_size=(2,2), padding='same', data_format='channels_last'),

进行数学以获取所需的输出。