层的不兼容输入（ndim=4，发现 ndim=3）答案

【问题标题】：Incompatible inputs of layers (ndim=4, found ndim=3)层的不兼容输入（ndim=4，发现 ndim=3）
【发布时间】：2020-06-16 00:45:17
【问题描述】：

我试图在 Keras 中重新创建 this Flower Recognition CNN。该模型似乎有效，至少在笔记本中（同时获得验证集的预测），但我需要在其他地方使用该模型。照片是 150x150，这就是我构建 CNN 的方式：

model = Sequential()
model.add(Conv2D(filters=32, kernel_size=(5,5), padding='Same', activation='relu', input_shape=(150, 150, 3)))
model.add(MaxPooling2D(pool_size=(2,2)))

model.add(Conv2D(filters=64, kernel_size=(3,3), padding='Same', activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

model.add(Conv2D(filters=96, kernel_size=(3,3), padding='Same', activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

model.add(Conv2D(filters=96, kernel_size=(3,3), padding='Same', activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2)))

model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dense(5, activation="softmax"))

不过，当我尝试使用加载的示例照片和模型在本地预测结果时：

model = model_from_json(open("model.json", "r").read())
model.load_weights('model.h5')

img = cv2.imread('/path/to/the/dir/testimage.jpg')
img = cv2.resize(img, (150, 150))
data = np.array(img)
model.summary()
result = model.predict(data)

我收到此错误：

ValueError: Input 0 of layer sequential_1 is incompatible with the layer: expected ndim=4, found ndim=3. Full shape received: [None, 150, 3]

模型汇总的输出为：

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 150, 150, 32)      2432      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 75, 75, 32)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 75, 75, 64)        18496     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 37, 37, 64)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 37, 37, 96)        55392     
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 18, 18, 96)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 18, 18, 96)        83040     
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 9, 9, 96)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 7776)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 512)               3981824   
_________________________________________________________________
activation_1 (Activation)    (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 5)                 2565      
=================================================================
Total params: 4,143,749
Trainable params: 4,143,749
Non-trainable params: 0

我不明白为什么它会有不同的形状。我已经读过 Reshape 可能会做这件事，但我正在使用 fit_generator 来适应训练集，我不确定这是否可行。

【问题讨论】：

标签： python tensorflow keras

【解决方案1】：

确保您的第一个维度为 1（要预测的 1 个图像）。您的模型期望 None 维度是样本数，2nd + 3rd 是图像分辨率，4th 是图像的 RGB 通道。

data = np.expand_dims(data, axis=0)

将为您的第一个轴添加一个额外的维度。

见：How can I add new dimensions to a Numpy array?

【讨论】：