【发布时间】：2019-12-09 14:15:20
【问题描述】：
#Now we will create a baseline CNN model similar to LeNet and below is utility class for BASELINE MODEL:
#train and val directory path:
train_dir = r'C:\Users\Ashu\Desktop\AAIC\Data\train'
val_dir = r'C:\Users\Ashu\Desktop\AAIC\Data\val'
nb_samples_train = 319835
nb_samples_val = 39995
class baseLINE_MODEL:
def __init__(self , img_width , img_height , epochs , batch_size):
self.img_width = img_width
self.img_height = img_height
self.epochs = epochs
self.batch_size = batch_size
def LENET(self):
'''The below architecture is almost similar to the LeNet with some small
modifications.The below model will be used as the baseline model only.
'''
if K.image_data_format == 'channel_first':
input_shape = (1 , 512 , 512)
else:
input_shape = (512 , 512 , 1)
self.model = Sequential(name="LENet")
self.model.add(Conv2D(filters=32 , kernel_size = (3,3) , input_shape = input_shape))
self.model.add(Activation('relu'))
self.model.add(MaxPooling2D(pool_size = (2,2) , strides=(2,2)))
self.model.add(Conv2D(filters = 62 , kernel_size = (3,3)))
self.model.add(Activation('relu'))
self.model.add(MaxPooling2D(pool_size = (2,2) , strides = (2,2)))
self.model.add(Conv2D(filters = 128 , kernel_size = (5,5)))
self.model.add(Activation('relu'))
self.model.add(MaxPooling2D(pool_size = (2 , 2) , strides = (2,2)))
self.model.add(Flatten())
self.model.add(Dense(100))
self.model.add(Activation('relu'))
self.model.add(Dense(16))
self.model.add(Activation('softmax'))
self.model.compile(loss = keras.losses.categorical_crossentropy , optimizer = 'adam' , metrics = ['accuracy'])
print(self.model.summary())
datagen = ImageDataGenerator(rescale = 1./255)
#the below code will create the train and validation generator.
#the images will be feed from the train and validation directory and get rescaled.
print('Feeding the train and val data.........')
train_generator = datagen.flow_from_directory(
train_dir,
target_size=(self.img_width,self.img_height),
batch_size=50,
class_mode = 'categorical')
validation_generator = datagen.flow_from_directory(
val_dir,
target_size=(self.img_width , self.img_height),
batch_size = 50,
class_mode = 'categorical')
print('Training Starting.....')
self.model.fit_generator(train_generator ,
samples_per_epoch = 319836 ,
epochs = self.epochs,
validation_data = validation_generator ,
validation_steps = 39995)
self.model.save_weights('fc_top_model.h5')
print('weights saved...')
我有 16 个类数据，并将图像放在 train 文件夹中，每个类有 16 个文件夹。这是为所有测试和验证完成的。但是当我尝试运行我的代码时，会在行中抛出错误： model.fit_generator() 输出：
_________________________________________________________________
activation_68 (Activation) (None, 122, 122, 128) 0
_________________________________________________________________
max_pooling2d_42 (MaxPooling (None, 61, 61, 128) 0
_________________________________________________________________
flatten_14 (Flatten) (None, 476288) 0
_________________________________________________________________
dense_27 (Dense) (None, 100) 47628900
_________________________________________________________________
activation_69 (Activation) (None, 100) 0
_________________________________________________________________
dense_28 (Dense) (None, 16) 1616
_________________________________________________________________
activation_70 (Activation) (None, 16) 0
=================================================================
Total params: 47,847,282
Trainable params: 47,847,282
Non-trainable params: 0
_________________________________________________________________
None
Feeding the train and val data.........
Found 319835 images belonging to 16 classes.
Found 39995 images belonging to 16 classes.
Training Starting.....
Epoch 1/1
---------------------------------------------------------------------------
OSError Traceback (most recent call last)
<ipython-input-45-0229da3ad25e> in <module>
----> 1 baseline.LENET()
错误：
et_size, interpolation)
117 elif color_mode == 'rgb':
118 if img.mode != 'RGB':
--> 119 img = img.convert('RGB')
120 else:
121 raise ValueError('color_mode must be "grayscale", "rgb", or "rgba"')
~\Miniconda3\envs\tensorflow\lib\site-packages\PIL\Image.py in convert(self, mode, matrix, dither, palette, colors)
928 """
929
--> 930 self.load()
931
932 if not mode and self.mode == "P":
~\Miniconda3\envs\tensorflow\lib\site-packages\PIL\TiffImagePlugin.py in load(self)
1095 def load(self):
1096 if self.use_load_libtiff:
-> 1097 return self._load_libtiff()
1098 return super(TiffImageFile, self).load()
1099
~\Miniconda3\envs\tensorflow\lib\site-packages\PIL\TiffImagePlugin.py in _load_libtiff(self)
1207
1208 if err < 0:
-> 1209 raise IOError(err)
1210
1211 return Image.Image.load(self)
OSError: -2
请告诉我这里有什么问题。我已经搜索了整个互联网但无法解决这个问题:(
P.S : 数据集是 RVL-CDIP 数据集。上面的代码在我使用猫狗数据集时运行良好。请帮助我。
【问题讨论】：
标签： image-processing keras deep-learning conv-neural-network