如何解决“输入形状的预期轴 -1 的值为 1，但收到的输入形状为 [None, 256, 256, 3]'”错误？

Question

I am trying to use a keras model. I trained model and want to use it from webcam. However, as far as I understand, the inputs I use while training the model and the inputs I receive from the camera do not match. How do I fix this problem?

Here code for train:

from keras.preprocessing.image import ImageDataGenerator
from keras.layers import Conv2D
from keras.layers import MaxPooling2D
from keras.layers import Dropout
from keras.layers import Dense
from keras.layers import Flatten

from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.models import Sequential, load_model
import tensorflow as tf
import numpy as np
import os

# plot pretty figures
import matplotlib
import matplotlib.pyplot as plt

plt.rcParams['axes.labelsize'] = 14
plt.rcParams['xtick.labelsize'] = 12
plt.rcParams['ytick.labelsize'] = 12

nbatch=32

train_datagen = ImageDataGenerator ( rescale=1./255,
                                     rotation_range=12.,
                                     width_shift_range=0.2,
                                     height_shift_range=0.2,
                                     zoom_range=0.15,
                                     horizontal_flip=True)

test_datagen = ImageDataGenerator (rescale=1./255)

train_gen = train_datagen.flow_from_directory(
    'images/train/',
    target_size=(256,256),
    color_mode='grayscale',
    batch_size=nbatch,
    classes=['NONE','ONE','TWO','THREE','FOUR','FIVE'],
    class_mode='categorical'
)

test_gen = test_datagen.flow_from_directory(
    'images/test/',
    target_size=(256,256),
    color_mode='grayscale',
    batch_size=nbatch,
    classes=['NONE','ONE','TWO','THREE','FOUR','FIVE'],
    class_mode='categorical'
)

for X, y in train_gen:
    print(X.shape, y.shape)

    plt.figure(figsize=(16,16))
    for i in range(25):
        plt.subplot(5,5,i+1)
        plt.axis('off')
        plt.title('Label: {}'.format(np.argmax(y[i])))
        img= np.uint8(255*X[i,:,:,0])
        plt.imshow(img,cmap='gray')
    break

plt.show()

model = Sequential()
model.add(Conv2D(32,(3,3),activation='relu',input_shape=(256,256,1)))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(128,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(256,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))
model.add(Flatten())
model.add(Dense(150, activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(6,activation='softmax'))

model.summary()

model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['acc'])

callback_list=[EarlyStopping(monitor='val_loss',patience=10),
               ModelCheckpoint(filepath='model_6cat_2.h6',monitor='val_loss',save_best_only=True),]

os.environ["CUDA_VISIBLE_DEVİCES"] = "0"
with tf.device('/GPU:0'):
    history = model.fit_generator(
        train_gen,
        steps_per_epoch=64,
        epochs=200,
        validation_data=test_gen,
        validation_steps=28,
        callbacks=callback_list
    )

plt.figure(figsize=(16,6))
plt.subplot(1,2,1)
nepochs=len(history.history['loss'])
plt.plot(range(nepochs),history.history['loss'], 'g-', label='train')
plt.plot(range(nepochs),history.history['val_loss'], 'c-', label='test')
plt.legend(prop={'size':20})
plt.ylabel('loss')
plt.xlabel('number of epochs')
plt.subplot(1,2,2)
plt.plot(range(nepochs),history.history['acc'], 'g-', label='train')
plt.plot(range(nepochs),history.history['val_acc'], 'c-', label='test')
plt.legend(prop={'size':20})
plt.ylabel('accuracy')
plt.xlabel('number of epochs')


X_test, y_test= [], []
for ibatch, (X,y) in enumerate(test_gen):
    X_test.append(X)
    y_test.append(y)
    ibatch+=1
    if (ibatch==5*28):break

X_test = np.concatenate(X_test)
y_test = np.concatenate(y_test)
y_test = np.int32([np.argmax(r) for  r in y_test])


y_pred = np.int32([np.argmax(r) for  r in model.predict(X_test)])
match=(y_test == y_pred)
print(("Testing Accuracy = {}").format(np.sum(match)*100/match.shape[0]))

Here code for predict:

model = load_model("C://Users//90544//OneDrive//Masaüstü//Yusuf// 
ödevler//kerasGiris//model_6cat_2.h6", compile = True)

  cap = cv2.VideoCapture(0)
  while 1:
    ret, frame = cap.read()
    if ret:
        frame = cv2.flip(frame, 1)
        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frame = cv2.resize(frame, (256, 256))
        frameNp = image.img_to_array(frame)
        frameNp = np.expand_dims(frameNp, axis=0)

        predictions = model.predict(frameNp)
        print(predictions)

        cv2.imshow("frame", frameNp)

        k = cv2.waitKey(1) & 0xff
        if k == 27: break  # ESC pressed

    cap.release()
    cv2.destroyAllWindows()

ValueError: Input 0 of layer sequential is incompatible with the layer: expected axis -1 of input shape to have value 1 but received input with shape [None, 256, 256, 3]

I tried to change the shape of the image I got from the camera, but I couldn't decide what the dimensions should be.

Answer 1

You can tell the input shape of a model from this line:

model.add(Conv2D(32,(3,3),activation='relu',input_shape=(256,256,1)))

This line means the model takes an image of shape (256, 256, 1) based on the input_shape argument, therefore the model expects to take images of that size.

Your error message means that you used an image of shape (256, 256, 3) and it expected 1 instead of 3 so you need to make the channels value to be 1 as in gray scale image instead of 3 which is BGR.

Add this line after the first line in the while loop in the code for predict:

frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

This line changes the channels of the image from BGR to Gray scale so as to match the input size channels for the model and the input shape requested by the model.