无法识别导致错误的图像-PIL.UnidentifiedImageError:无法识别图像文件

Question

在训练NeuralNetwork模型时，我得到以下错误：

PIL.UnidentifiedImageError: cannot identify image file <_io.BytesIO object at 0x7fd6342bdfb0>

我已将PIL版本更新为最新版本。我无法指出是哪个图像文件导致了问题。所有使用的图像文件都是JPG文件。

下面是我正在使用的代码。

from tensorflow.keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255,
shear_range = 0.2,
zoom_range = 0.2,
horizontal_flip = True
)
test_datagen = ImageDataGenerator(rescale = 1./255)
training_set = train_datagen.flow_from_directory(IMG_PATH_TRAIN,
target_size = (224,224),
batch_size = 32,
shuffle = False
#class_mode = 'categorical'
)
test_set = test_datagen.flow_from_directory(IMG_PATH_TEST,
target_size = (224,224),
batch_size = 32,
shuffle = False
#class_mode = 'categorical'
)
lr = 0.003
sgd = SGD(lr=lr, momentum=0.9, nesterov=False)
adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)
input_shape = (224,224,3)
inputs = Input(input_shape)
def set_model(base_model):
x =  base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(82, activation='softmax')(x)
base_model = Model(inputs=inputs, outputs= [x])
batch_size = 32

for layer in base_model.layers:
layer.trainable = True

base_model.compile(loss=['categorical_crossentropy','categorical_crossentropy','categorical_crossentropy'],
loss_weights=[1,1,1],
optimizer= sgd,
metrics=['accuracy']) #, 'top_k_categorical_accuracy'])

base_model.fit(training_set,
steps_per_epoch=20991//batch_size,
epochs=1, 
validation_data=test_set, 
validation_steps=7456//batch_size, #test_samples // batch_size,
verbose=1)
return base_model
base_model_inception = InceptionV3(include_top=False, weights=None, input_tensor = inputs)
base_model_inception = set_model(base_model_inception)

Answer 1

您检查了图像的大小和格式吗？有些图像可能没有正确格式化，例如RGB与RGBA。

此外，如果图像是以gif格式读取的，即p格式，那么TensorFlow库将无法在数据生成过程中进行转换，这可能会导致错误。