训练CNN识别验证码时出现输入形状不兼容错误求助

我尝试用简单CNN识别验证码里的数字，但运行model.fit时出错：

ValueError: Input 0 of layer "sequential" is incompatible with the layer: expected shape=(None, 38, 321, 3), found shape=(None, 321, 38, 3)

代码文件（python.py） #

IMAGEPATH = 'captcha'
dirs = os.listdir(IMAGEPATH)
  
images=[]
test_images=[]
train_labels=[]
train_images, train_labels = [],[]
test_labels=[]

X=[]
Y=[]

w=38 
h=321 
i=0

for name in dirs:
    file_paths = glob.glob(path.join(IMAGEPATH+"/"+name, '*.*'))
    for path3 in file_paths:
        try:  
            img = cv2.imread(path3)
            img = cv2.resize(img, (w, h))
            im_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
            ret, im_res = cv2.threshold(im_rgb,180,255,cv2.THRESH_BINARY)
            if img is not None:
                images.append(im_res)
                test_images.append(im_res)
                name=path3.split(os.path.sep)[-1]
                getdata=(name.split('.')[-2])
                labels=getdata.split('-')[-1]
                train_labels.append(labels)
                #train_labels.append(i)
            i=i+1
        except:
            print(os.path.join(file_paths,name),"error!")
            pass

print(len(images),len(train_labels))

images=np.array(images)
train_labels==np.array(train_labels)

(X_train, X_test, Y_train, Y_test) = train_test_split(images, train_labels, test_size=0.2, random_state=0) 

X_train_normalize=X_train.reshape(X_train.shape).astype("float")/255.0
X_test_normalize=X_test.reshape(X_test.shape).astype("float")/255.0

lb = LabelBinarizer().fit(Y_train)
Y_train_OneHot = lb.transform(Y_train)
Y_test_OneHot = lb.transform(Y_test)

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(w,h,3)))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(64, (3, 3), activation='relu'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(64, (3, 3), activation='relu'))

model.summary()

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

history = model.fit(X_train_normalize,
                              epochs=10
                              )

验证码文件路径 #

captcha\0A3F0A\teacher-26.jpg
captcha\1673A2\teacher-24.jpg
captcha\1E82F1\teacher-25.jpg
captcha\33C958\teacher-20.jpg
captcha\33DC34\teacher-17.jpg
captcha\3AF35E\teacher-14.jpg
captcha\3B1C9E\teacher-12.jpg
captcha\4207C9\teacher-9.jpg
captcha\4B5AC4\teacher-21.jpg
captcha\4DD685\teacher-3.jpg
captcha\4E44F0\teacher-16.jpg
captcha\6BEBFE\teacher-11.jpg
captcha\6DCE49\teacher-15.jpg
captcha\6E16E9\teacher-4.jpg
captcha\c8BA540\teacher-8.jpg
captcha\949BA9\teacher-23.jpg
captcha\99F671\teacher-7.jpg
captcha\A1564D\teacher-10.jpg
captcha\A4B883\teacher-13.jpg
captcha\D27153\teacher-6.jpg
captcha\D36E65\teacher-19.jpg
captcha\DC602A\teacher-22.jpg
captcha\DDE328\teacher-18.jpg
captcha\DED2FB\teacher-2.jpg

完整报错信息 #

d:\OpecvImage\lookthree.py:83: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
  history = model.fit_generator(X_train_normalize,
Traceback (most recent call last):
  File "d:\OpecvImage\lookthree.py", line 83, in <module>
    history = model.fit_generator(X_train_normalize,
  File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\engine\training.py", line 2209, in fit_generator
    return self.fit(
  File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\utils\traceback_utils.py", line 67, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\tensorflow\python\framework\func_graph.py", line 1147, in autograph_handler
    raise e.ag_error_metadata.to_exception(e)
ValueError: in user code:

    File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\engine\training.py", line 1021, in train_function  *
        return step_function(self, iterator)
    File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\engine\training.py", line 1010, in step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\engine\training.py", line 1000, in run_step  **
        outputs = model.train_step(data)
    File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\engine\training.py", line 859, in train_step
        y_pred = self(x, training=True)
    File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\utils\traceback_utils.py", line 67, in error_handler
        raise e.with_traceback(filtered_tb) from None
    File "C:\Users\user\AppData\Local\Programs\Python\Python39\lib\site-packages\keras\engine\input_spec.py", line 264, in assert_input_compatibility
        raise ValueError(f'Input {input_index} of layer "{layer_name}" is '

    ValueError: Input 0 of layer "sequential" is incompatible with the layer: expected shape=(None, 38, 321, 3), found shape=(None, 321, 38, 3)

核心问题：图像宽高顺序不匹配 #

你定义了w=38，h=321，并在模型输入层指定input_shape=(w,h,3)，也就是期望图像形状为**(38, 321, 3)（高度38，宽度321，3通道）。但OpenCV的resize函数参数顺序是(width, height)，你调用cv2.resize(img, (w, h))实际把图像改成了宽38，高321**，对应数组形状为(321, 38, 3)，和模型期望的顺序完全相反，导致形状不匹配。

修复步骤： #

1. 调整resize参数顺序：将cv2.resize(img, (w, h))改为cv2.resize(img, (h, w))，确保输出图像的宽高对应模型要求的(h, w)，最终数组形状为(38, 321, 3)。

2. 修正标签转换代码：train_labels==np.array(train_labels)是无效的比较操作，改为train_labels = np.array(train_labels)才能将标签转换为NumPy数组。

3. 补充fit的标签参数：当前model.fit只传了训练数据，缺少标签输入。结合你使用的SparseCategoricalCrossentropy损失函数，需要传入原始整数标签Y_train，而非独热编码的Y_train_OneHot。

修正后的关键代码片段： #

# 修正resize参数顺序
img = cv2.resize(img, (h, w))

# 修正标签转换
train_labels = np.array(train_labels)

# 修正model.fit调用
history = model.fit(X_train_normalize, Y_train, epochs=10)

内容的提问来源于stack exchange，提问作者Paul