如何通过表格和图表图像制作下面描述的 CNN 网络？

Question

我将一个 CNN 网络描述为一篇研究论文，请告诉我在哪里做错了？

因为它显示以下错误：

ValueError：对于输入形状为 [?,5,5,8] 的“max_pooling2d_1/MaxPool”（操作：“MaxPool”），从 5 中减去 68 导致的负尺寸尺寸：[?,5,5,8]。

以下图片中提供了说明：

第一幅图像由卷积和Max Pooling细节描述，第二幅图像描述跟随框图。

import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
import numpy as np
import cv2
import os

path1="/home/sanjay/CASIA_B90PerfectCentrallyAlinged_Resized_with_140by140_Energy_Image/"
all_images = []
all_labels = []
subjects = os.listdir(path1)
numberOfSubject = len(subjects)
print('Number of Subjects: ', numberOfSubject)
for number1 in range(0, numberOfSubject):  # numberOfSubject
    path2 = (path1 + subjects[number1] + '/')
    sequences = os.listdir(path2);
    numberOfsequences = len(sequences)
    for number2 in range(4, numberOfsequences):
        path3 = path2 + sequences[number2]
        img = cv2.imread(path3 , 0)
        img = img.reshape(140, 140, 1)
        all_images.append(img)
        all_labels.append(number1)
x_train = np.array(all_images)
y_train = np.array(all_labels)
y_train = keras.utils.to_categorical(y_train)
print(y_train.shape)
print(x_train.shape)

all_images = []
all_labels = []
for number1 in range(0, numberOfSubject):  # numberOfSubject
    path2 = (path1 + subjects[number1] + '/')
    sequences = os.listdir(path2);
    numberOfsequences = len(sequences)
    for number2 in range(0, 4):
        path3 = path2 + sequences[number2]
        img = cv2.imread(path3 , 0)
        img = img.reshape(140, 140, 1)
        all_images.append(img)
        all_labels.append(number1)
x_test = np.array(all_images)
y_test = np.array(all_labels)
y_test = keras.utils.to_categorical(y_test)

print(y_test.shape)
print(x_test.shape)
#print(y_test)

batch_size =123
num_classes = 123
epochs = 80

model = Sequential()
model.add(Conv2D(8, kernel_size=(136,136), activation='tanh', input_shape=(140,140,1)))
model.add(MaxPooling2D(pool_size=(68, 68)))
model.add(Conv2D(8, kernel_size=64, activation='tanh'))
model.add(MaxPooling2D(pool_size=(32, 32)))
model.add(Conv2D(8, kernel_size=28, activation='tanh'))
model.add(MaxPooling2D(pool_size=(14, 14)))
model.add(Conv2D(8, kernel_size=10, activation='tanh'))
model.add(MaxPooling2D(pool_size=(5, 5)))
model.add(Flatten())
model.add(Dense(1000, activation='tanh'))
model.add(Dense(123, activation='softmax'))

model.compile(loss=keras.losses.categorical_crossentropy, optimizer=keras.optimizers.Adadelta(), metrics=['accuracy'])

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1, validation_data=(x_test, y_test))

score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

这里我有 CASIA_B 数据集的 123 个主题，每个类有 10 帧。

Answer 1

发生错误是因为没有足够的信息来了解如何设置卷积层的内核大小以及最大池化的工作方式。我建议请查看this，在那里您可以找到有关卷积以及如何设置内核大小的详细信息。也适用于pooling layer。

对于您的实施，

model = Sequential()
model.add(Conv2D(8, kernel_size=(5,5), activation='tanh', input_shape=(140,140,1)))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(8, kernel_size=(5,5), activation='tanh'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(8, kernel_size=(5,5), activation='tanh'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(8, kernel_size=(5,5), activation='tanh'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Flatten())
model.add(Dense(1000, activation='tanh'))
model.add(Dense(123, activation='softmax'))
model.summary()

Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 136, 136, 8)       208       
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 68, 68, 8)         0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 64, 64, 8)         1608      
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 32, 32, 8)         0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 28, 28, 8)         1608      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 14, 14, 8)         0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 10, 10, 8)         1608      
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 5, 5, 8)           0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 200)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 1000)              201000    
_________________________________________________________________
dense_2 (Dense)              (None, 123)               123123    
=================================================================
Total params: 329,155
Trainable params: 329,155
Non-trainable params: 0
_________________________________________________________________

更新

您的实现是必需的自定义层，您可以在this github repo 中看到它。我不确定它是否已经完全开发。

你需要下载this文件或者可以克隆完整的存储库并像这样导入，

from Conv2D121 import Conv2D121


model = Sequential()
model.add(Conv2D(8, (5, 5), padding='valid',
                 input_shape=(140, 140, 1)))
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(2,2), strides=2, padding='valid'))
model.add(Conv2D121(8, (5, 5), padding='valid'))
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(2,2), strides=2, padding='valid'))
model.add(Conv2D121(8, (5, 5), padding='valid'))
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(2,2), strides=2, padding='valid'))
model.add(Conv2D121(8, (5, 5), padding='valid'))
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(2,2), strides=2, padding='valid'))
model.add(Flatten())
model.add(Dense(1000, activation='tanh'))
model.add(Dense(123, activation='softmax'))
model.summary()