三维卷积神经网络输入形状

Question

我在使用Keras和Python对3D形状进行分类时遇到了一个问题。我有一个文件夹，里面有一些JSON格式的模型。我将这些模型读入Numpy数组。模型是25*25*25，表示体素化模型的占用网格（每个位置表示位置（i、j、k）中的体素是否有点），因此我只有1个输入通道，就像2D图像中的灰度图像一样。我拥有的代码如下：

import numpy as np
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution3D, MaxPooling3D
from keras.optimizers import SGD
from keras.utils import np_utils
from keras import backend as K

# Number of Classes and Epochs of Training
nb_classes = 3 # cube, cone or sphere
nb_epoch = 100
batch_size = 2

# Input Image Dimensions
img_rows, img_cols, img_depth = 25, 25, 25

# Number of Convolutional Filters to use
nb_filters = 32

# Convolution Kernel Size
kernel_size = [5,5,5]

X_train, Y_train = [], []

# Read from File
import os
import json

i=0
for filename in os.listdir(os.path.join(os.getcwd(), 'models')):
    with open(os.path.join(os.getcwd(), 'models', filename)) as f:
        file = f.readlines()
        json_file = '\n'.join(file)
        content = json.loads(json_file)
        occupancy = content['model']['occupancy']
        form = []
        for value in occupancy:
            form.append(int(value))
        final_model = [ [ [ 0 for i in range(img_rows) ]
                              for j in range(img_cols) ]
                              for k in range(img_depth) ]
        a = 0
        for i in range(img_rows):
            for j in range(img_cols):
                for k in range(img_depth):
                    final_model[i][j][k] = form[a]
                    a = a + 1
        X_train.append(final_model)
        Y_train.append(content['model']['label'])

X_train = np.array(X_train)
Y_train = np.array(Y_train)

# (1 channel, 25 rows, 25 cols, 25 of depth)
input_shape = (1, img_rows, img_cols, img_depth)

# Init
model = Sequential()

# 3D Convolution layer
model.add(Convolution3D(nb_filters, kernel_size[0], kernel_size[1], kernel_size[2],
                        input_shape=input_shape,
                        activation='relu'))

# Fully Connected layer
model.add(Flatten())
model.add(Dense(128,
          init='normal',
          activation='relu'))
model.add(Dropout(0.5))

# Softmax Layer
model.add(Dense(nb_classes,
                init='normal'))
model.add(Activation('softmax'))

# Compile
model.compile(loss='categorical_crossentropy',
              optimizer=SGD())

# Fit network
model.fit(X_train, Y_train, nb_epoch=nb_epoch,
         verbose=1)

在此之后，我得到以下错误

使用TensorFlow后端。回溯（最后一次调用）：文件“/usr/local/lib/python3.6/site packages/tensorflow/python/framework/common\u shapes.py”，第670行，在\u call\u cpp\u shape\u fn\u impl status）文件“/usr/local/cillar/python3/3.6.0/Frameworks/python.framework/Versions/3.6/lib/python3.6/contextlib.py”第89行，在exit next（self.gen）文件“/usr/local/lib/python3.6/site packages/tensorflow/python/framework/errors\u impl.py”的第469行中，在raise\u exception\u not\u ok\u status pywrap\u tensorflow中。TF\u GetCode（状态））tensorflow。蟒蛇框架错误\u impl。InvalidArgumentError：输入形状为[？，1,25,25,25]，[5,5,25,32]的“Conv3D”（op：“Conv3D”）从1中减去5，导致尺寸为负。

在处理上述异常时，发生了另一个异常：

回溯（最后一次调用）：文件“CNN\u 3D.py”，第76行，在activation='relu'）中；文件“/usr/local/lib/python3.6/site packages/keras/models.py”，第299行，在add layer中。在调用self中的create\u input\u layer self（x）文件“/usr/local/lib/python3.6/site packages/keras/engine/topology.py”中的第401行中创建\u input\u layer（batch\u input\u shape，input\u dtype）文件“/usr/local/lib/python3.6/site packages/keras/engine/topology.py”，第572行。在add\u inbound\u节点中添加\u inbound\u节点（inbound\u layers、node\u index、tensor\u index）文件“/usr/local/lib/python3.6/site packages/keras/engine/topology.py”，第635行。创建\u节点（self、inbound\u layers、node\u index、tensor\u index）文件“/usr/local/lib/python3.6/site packages/keras/engine/topology.py”，第166行，在create\u node output\u tensors=to\u list（outbound\u layer.call（input\u tensors[0]，mask=input\u masks[0]）文件“/usr/local/lib/python3.6/site packages/keras/layers/coulsional.py”，第1234行，在call filter\u shape=self中。W\u shape）File“/usr/local/lib/python3.6/site packages/keras/backend/tensorflow\u backend.py”，第2831行，conv3d x=tf。nn。conv3d（x，kernel，strips，padding）文件“/usr/local/lib/python3.6/site-packages/tensorflow/python/ops/gen\u nn\u ops.py”，第522行，在conv3d-strips=strips，padding=padding，name=name）文件“/usr/local/lib/python3.6/site-packages/tensorflow/python/framework/op\u def\u library.py”，第763行，在apply\u-op\u-def）文件中python/framework/ops.py“，第2397行，在create\u op set\u shapes\u for\u outputs（ret）文件“/usr/local/lib/python3.6/site packages/tensorflow/python/framework/ops.py”中，第1757行，在set\u shapes\u for\u outputs shapes=shape\u func（op）文件“/usr/local/lib/python3.6/site packages/tensorflow/python/framework/ops.py”中，第1707行，在call\u with\u required return call\u cpp\u shape\u fn（op，require\u shape\u fn=True）File“/usr/local/lib/python3.6/site packages/tensorflow/python/framework/common\u shapes.py”中，第610行，在call\u cpp\u shape\u fn debug\u python\u shape\u fn，require\u shape\u fn）File“/usr/local/lib/python3.6/site packages/tensorflow/python/framework/common\u shapes.py”中，第675行，在\u call\u cpp\u shape\u fn\u impl raise ValueError（错误消息）ValueError：输入形状为[？，1,25,25,25]，[5,5,25,32]的“Conv3D”（op：“Conv3D”）从1中减去5导致尺寸为负。

我做错了什么才得到这个错误？

Answer 1

我认为问题在于，您正在No ordering中设置输入形状，但您正在将Keras与Tensorflow backend和Tensorflow img ordering一起使用。此外，必须将y\U列数组转换为类别标签。

更新的代码：

from keras.utils import np_utils
from keras import backend as K

if K.image_dim_ordering() == 'th':
    X_train = X_train.reshape(X_train.shape[0], 1, img_rows, img_cols, img_depth)
    input_shape = (1, img_rows, img_cols, img_depth)
else:
    X_train = X_train.reshape(X_train.shape[0], img_rows, img_cols, img_depth, 1)
    input_shape = (img_rows, img_cols, img_depth, 1)

Y_train = np_utils.to_categorical(Y_train, nb_classes)

添加此行应该可以修复它。