我使用OpenCV从视频文件中加载帧到数组中，并使用sklearn将数据分割为X_train和X_test。

我的X_train.shape是(363, 1, 40, 40, 15)，目前我正在处理4个类别，我用来学习这些数据的模型代码如下：

    model = Sequential()    model.add(Conv3D(32, (3,3,3), activation='relu', input_shape=(1, 40, 40, 15), data_format='channels_first'))    model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))    model.add(Conv3D(64, (3,3,3), activation='relu'))    model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))    model.add(Conv3D(128, (3,3,3), activation='relu'))    model.add(Conv3D(128, (3,3,3), activation='relu'))    model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))    model.add(Conv3D(256, (2,2,2), activation='relu'))    model.add(Conv3D(256, (2,2,2), activation='relu'))    model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))    model.add(Flatten())    model.add(Dense(1024))    model.add(Dropout(0.5))    model.add(Dense(1024))    model.add(Dropout(0.5))    model.add(Dense(4, activation='softmax'))

当我尝试加载模型时，我遇到了以下错误：

ValueError: Negative dimension size caused by subtracting 2 from 1 for 'conv3d_44/convolution' (op: 'Conv3D') with input shapes: [?,25,1,1,256], [2,2,2,256,256].

有人能帮我吗？

回答：

这个问题在StackOverflow上已经被讨论过多次：请参见这里或这里。根据你的卷积和池化层参数，每次Conv3D和MaxPooling3D后张量都会被下采样。以下是模型在崩溃前的状态：

Layer (type)                 Output Shape              Param #   =================================================================conv3d_1 (Conv3D)            (None, 32, 38, 38, 13)    896       _________________________________________________________________max_pooling3d_1 (MaxPooling3 (None, 32, 19, 19, 13)    0         _________________________________________________________________conv3d_2 (Conv3D)            (None, 30, 17, 17, 64)    22528     _________________________________________________________________max_pooling3d_2 (MaxPooling3 (None, 30, 8, 8, 64)      0         _________________________________________________________________conv3d_3 (Conv3D)            (None, 28, 6, 6, 128)     221312    _________________________________________________________________conv3d_4 (Conv3D)            (None, 26, 4, 4, 128)     442496    _________________________________________________________________max_pooling3d_3 (MaxPooling3 (None, 26, 2, 2, 128)     0         _________________________________________________________________conv3d_5 (Conv3D)            (None, 25, 1, 1, 256)     262400    

张量(None, 25, 1, 1, 256)无法进一步下采样，因此出现了错误。

解决方案是调整Conv3D参数：要么使用padding='same'（在这种情况下，卷积后张量形状保持不变，仅在池化层后减半），要么将滤波器大小从3减小到2。

示例：

model = Sequential()model.add(Conv3D(32, (3,3,3), activation='relu', input_shape=(1, 40, 40, 15), data_format='channels_first', padding='same'))model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))model.add(Conv3D(64, (3,3,3), activation='relu', padding='same'))model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))model.add(Conv3D(128, (3,3,3), activation='relu', padding='same'))model.add(Conv3D(128, (3,3,3), activation='relu', padding='same'))model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))model.add(Conv3D(256, (2,2,2), activation='relu', padding='same'))model.add(Conv3D(256, (2,2,2), activation='relu', padding='same'))model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2)))model.add(Flatten())model.add(Dense(1024))model.add(Dropout(0.5))model.add(Dense(1024))model.add(Dropout(0.5))model.add(Dense(4, activation='softmax'))model.summary()