我正在尝试使用training_label.CSV和两个文件夹（该模型的（测试/培训）文件夹）来识别语音声谱图中的重音。
我正在运行模型，但是遇到错误：
ValueError: Error when checking input: expected input_2 to have shape (128, 173, 1) but got array with shape (128, 173, 3)
我不确定要进行哪些更改才能使其正常工作？
# Importing the Keras libraries and packages
import pandas as pd
import numpy as np
from skimage.io import imread
from sklearn.model_selection import train_test_split
from matplotlib import pyplot as plt
%matplotlib inline
import keras
from keras.models import Sequential
from keras import optimizers
from keras.utils import to_categorical
from keras.layers import Input, Dense, Flatten, Dropout, Conv2D, MaxPooling2D, GlobalAveragePooling2D, Activation, concatenate
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.normalization import BatchNormalization
from keras.models import Model
from keras.callbacks import ModelCheckpoint, ReduceLROnPlateau
from keras.preprocessing import image
# Initialising the CNN
classifier = Sequential()
# Step 1 - Convolution
classifier.add(Conv2D(32, (3, 3), input_shape = (128, 173, 1), activation = 'relu'))
# Step 2 - Pooling
classifier.add(MaxPooling2D(pool_size = (2, 2)))
# Adding a second convolutional layer
classifier.add(Conv2D(32, (3, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
# Step 3 - Flattening
classifier.add(Flatten())
# Step 4 - Full connection
classifier.add(Dense(units = 128, activation = 'relu'))
classifier.add(Dense(units = 1, activation = 'sigmoid'))
# Compiling the CNN
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])
#Read Data
train_df = pd.read_csv('dataset/training_set_labels.csv', index_col=0)
train_df['Canada_file'] = train_df.index.map(lambda id: f'dataset/training/{id}.png')
train_df['India_file'] = train_df.index.map(lambda id: f'dataset/training/{id}.png')
train_df['England_file'] = train_df.index.map(lambda id: f'dataset/training/{id}.png')
print(train_df.head())
#Read Spectrograms
def read_spectograms(file_paths, img_rows, img_cols, as_gray, channels):
"""
Reads the spectogram files from disk and normalizes the pixel values
@params:
file_paths - Array of file paths to read from
img_rows - The image height.
img_cols - The image width.
as_grey - Read the image as Greyscale or RGB.
channels - Number of channels.
@returns:
The created and compiled model (Model)
"""
images = []
for file_path in file_paths:
images.append(imread(file_path, as_grey = as_gray))
images = np.asarray(images, dtype=np.float32)
# normalize
images = images / np.max(images)
# reshape to match Keras expectaions
images = images.reshape(images.shape[0], img_rows, img_cols, channels)
return images
#Parameters
as_gray = True
in_channel = 4
if as_gray:
in_channel = 1
img_rows, img_cols = 128, 173
num_classes = 3 # number of accents
batch_size = 32
epochs = 1
input_shape = (img_rows, img_cols, in_channel)
input_img = Input(shape = input_shape)
#Accent Files
x_train_Canada = read_spectograms(train_df.Canada_file.values, img_rows, img_cols, as_gray, in_channel)
x_train_India = read_spectograms(train_df.India_file.values, img_rows, img_cols, as_gray, in_channel)
x_train_England = read_spectograms(train_df.England_file.values, img_rows, img_cols, as_gray, in_channel)
#Labels
labels = train_df.Accent.values
# convert class vectors to binary class matrices One Hot Encoding
labels = keras.utils.to_categorical(labels, num_classes)
#Show Data
Accent = [
'Canada',
'India',
'England']
# pick a random index from the list
rn_Accent = np.random.choice(train_df.Accent.values)
rn_label = train_df.Accent.values[rn_Accent]
rn_Canada = x_train_Canada[rn_Accent]
rn_India = x_train_India[rn_Accent]
rn_England = x_train_England[rn_Accent]
plt.figure()
plt.axis('off')
plt.suptitle(f"{Accent[rn_label]} (Label: {rn_label})", fontsize="x-large")
plt.subplot(121)
curr_img = None
if as_gray:
curr_img = np.reshape(rn_Canada, (img_rows, img_cols))
else:
curr_img = np.reshape(rn_Canada, (img_rows, img_cols, in_channel))
plt.imshow(curr_img, cmap='gray')
plt.title("Canada")
plt.xticks([])
plt.yticks([])
plt.subplot(122)
if as_gray:
curr_img = np.reshape(rn_India, (img_rows, img_cols))
else:
curr_img = np.reshape(rn_India, (img_rows, img_cols, in_channel))
plt.imshow(curr_img, cmap='gray')
plt.title("India")
plt.xticks([])
plt.yticks([])
plt.subplot(123)
if as_gray:
curr_img = np.reshape(rn_England, (img_rows, img_cols))
else:
curr_img = np.reshape(rn_England, (img_rows, img_cols, in_channel))
plt.imshow(curr_img, cmap='gray')
plt.title("England")
plt.xticks([])
plt.yticks([])
plt.show()
#Split in Train and Test Batches
x_train_comp = np.stack((x_train_Canada, x_train_India, x_train_England), axis=4)
x_train, x_test, y_train, y_test = train_test_split(x_train_comp, labels, test_size = 0.3, random_state=666)
# take them apart
x_train_Canada = x_train[:,:,:,0]
x_test_Canada = x_test[:,:,:,0]
x_train_India = x_train[:,:,:,0]
x_test_India = x_test[:,:,:,0]
x_train_England = x_train[:,:,:,0]
x_test_England = x_test[:,:,:,0]
#Create the Model
def create_convolution_layers(input_img):
model = Conv2D(32, (3, 3), padding='same', input_shape=input_shape)(input_img)
model = LeakyReLU(alpha=0.1)(model)
model = MaxPooling2D((2, 2),padding='same')(model)
model = Dropout(0.25)(model)
model = Conv2D(64, (3, 3), padding='same')(model)
model = LeakyReLU(alpha=0.1)(model)
model = MaxPooling2D(pool_size=(2, 2),padding='same')(model)
model = Dropout(0.25)(model)
model = Conv2D(128, (3, 3), padding='same')(model)
model = LeakyReLU(alpha=0.1)(model)
model = MaxPooling2D(pool_size=(2, 2),padding='same')(model)
model = Dropout(0.4)(model)
return model
Canada_input = Input(shape=input_shape)
Canada_model = create_convolution_layers(Canada_input)
India_input = Input(shape=input_shape)
India_model = create_convolution_layers(India_input)
England_input = Input(shape=input_shape)
England_model = create_convolution_layers(England_input)
conv = concatenate([Canada_model, India_model, England_model])
conv = Flatten()(conv)
dense = Dense(512)(conv)
dense = LeakyReLU(alpha=0.1)(dense)
dense = Dropout(0.5)(dense)
output = Dense(num_classes, activation='softmax')(dense)
model = Model(inputs=[Canada_input, India_input, England_input], outputs=[output])
opt = optimizers.Adam()
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=['accuracy'])
print(model.summary())
#Train the Model
best_weights_file="weights.best.hdf5"
checkpoint = ModelCheckpoint(best_weights_file, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
#lr_reducer = ReduceLROnPlateau(verbose=1)
callbacks = [checkpoint]
model.fit([x_train_Canada, x_train_India, x_train_England], y_train,
batch_size=batch_size,
epochs=epochs,
callbacks=callbacks,
verbose=1,
validation_data=([x_test_Canada, x_test_India, x_test_England], y_test),
shuffle=True)
#Basic Evaluation
# load weights
#model.load_weights(best_weights_file)
#final_loss, final_acc = model.evaluate([x_test_accent, x_test_accent], y_test, verbose=1)
#print("Final loss: {0:.6f}, final accuracy: {1:.6f}".format(final_loss, final_acc))
#Predict
predict_df = pd.read_csv('dataset/submission_format.csv', index_col=0)
predict_df['Canada_file'] = predict_df.index.map(lambda id: f'dataset/testing/{id}.png')
x_test_accent = read_spectograms(predict_df.Canada_file.values, img_rows, img_cols, as_gray, in_channel)
#Get the predictions for the test data
predicted_classes = model.predict([x_test_Canada, x_test_India, x_test_England])
predict_df.Accent = np.argmax(predicted_classes,axis=1)
#predict_df = predict_df.drop(['Accent_file''], axis=1)
predict_df.to_csv('submission.csv')
files.download('submission.csv')
我不确定为什么会出现此错误。我对Python非常陌生，这是我正在做的Microsoft课程的一部分。这是link to the course
我想要达到的最终结果是使用模型的Testing文件夹中的数据生成Submission_Format.csv，其中模型已经预测了每个频谱图.png文件的每个重音。
答案 0 :(得分：0)
您只需要将输入形状更改为（128，173，3），因为您正在处理包含3个不同颜色通道的图像