ValueError：CorrMM：不可能的输出形状

Question

我需要在具有65536行（代表每个图像1个图像），49列（7x7图像）和二进制类（第50列）的数据集上建立CNN模型。 （7800行测试集）

我参考使用mnist数据集执行CNN的示例，但我未能建立火车模型。

这里是我的代码：

from __future__ import print_function 
import keras 
from keras.models import Sequential 
from keras.layers import Dense, Dropout, Flatten 
from keras.layers import Conv2D, MaxPooling2D 
from keras.utils import np_utils 
from keras import backend as K 
K.set_image_dim_ordering('th') 
import pandas as pd 
import numpy as np 



# input image dimensions 
img_rows, img_cols = 7, 7 

# fix random seed for reproducibility 
seed = 7 
np.random.seed(seed) 

x_train = pd.read_csv('palm_3x3_test.csv') 
x_train.drop(['class'],axis=1,inplace=True) 

x_train = x_train.as_matrix().reshape(-1,7, 7) 

y_train = pd.read_csv('palm_3x3_test.csv') 
y_train = y_train[['class']] 


x_test = pd.read_csv('palm_3x3_data.csv') 
x_test.drop(['class'],axis=1,inplace=True) 
x_test = x_test.as_matrix().reshape(-1,7, 7) 

y_test = pd.read_csv('palm_3x3_data.csv') 
y_test = y_test[['class']] 


# reshape to be [samples][pixels][width][height] 
x_train_final = x_train.reshape(x_train.shape[0],1,7,7).astype('float32') 
x_test_final = x_test.reshape(x_test.shape[0],1,7, 7).astype('float32') 

print(x_train.shape) 
print(x_test.shape) 
print(x_train_final.shape) 
print(x_test_final.shape) 
print(y_train.shape) 
print(y_test.shape) 

# normalize inputs from 0-255 to 0-1 
x_train_final = x_train_final/255 
x_test_final = x_test_final/255 

# one hot encode outputs 
y_train = np_utils.to_categorical(y_train) 
y_test = np_utils.to_categorical(y_test) 
num_classes = y_test.shape[1] 

print (y_test.shape) 
print(y_train.shape) 

def baseline_model(): 
#  create model 

    model = Sequential() 
    model.add(Conv2D(30,(5,5), padding='valid', activation='relu',input_shape=(1,7,7))) 
    model.add(MaxPooling2D(pool_size=(2, 2))) 
    model.add(Conv2D(15 ,(3, 3), activation='relu')) 
    model.add(MaxPooling2D(pool_size=(2, 2))) 
    model.add(Dropout(0.2)) 
    model.add(Flatten()) 
    model.add(Dense(128, activation='relu')) 
    model.add(Dense(50, activation='relu')) 
    model.add(Dense(num_classes, activation='softmax')) 

    # Compile model 
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) 
    return model 

# build the model 
model = baseline_model() 
# Fit the model 
model.fit(x_train_final,y_train, validation_data=(x_test_final,y_test), nb_epoch=10, batch_size=200,verbose=2) 
# Final evaluation of the model 
scores = model.evaluate(x_test,y_test, verbose=0) 
print("CNN Error: %.2f%%" % (100-scores[1]*100)) 

我使用Theano和我的数据输出形状：

x_train.shape = (65536L, 7L, 7L) 
x_test.shape = (7800L, 7L, 7L) 
x_train_final.shape = (65536L, 1L, 7L, 7L) 
x_test_final.shape = (7800L, 1L, 7L, 7L) 
y_train.shape = (65536L, 2L) 
y_test.shape = (7800L, 2L) 

，而我运行它的错误提示：

ValueError: CorrMM: impossible output shape 
    bottom shape: 200 x 30 x 1 x 1 
    weights shape: 15 x 30 x 3 x 3 
    top shape: 200 x 15 x -1 x -1 

Answer 1

您对第二个卷积图层的输入小于您希望应用的滤镜。您的输入图像是(7,7);如果您使用valid填充与(5,5)筛选器，您将有一个筛选的图像是(3,3)。应用最大池后，您将剩下一个(1,1)图像，该图像与您的辅助卷积图层不兼容，该图像的滤镜尺寸为(3,3)。

编译模型后，您可以运行model.summary()来查看每个图层的输出形状。消极的形状导致你看到的impossible output shape错误。在第一卷积层中改变padding='valid'至padding='same'。这不会完全解决您的问题，您还需要更改第二个卷积图层的填充以确保完全连接的图层具有输出。考虑到图像的小尺寸，您可能需要重新考虑使用最大池。

This question是有效和相同填充之间差异的很好解释。