keras 将datagen.flow_from_directory与图像分割和类数结合使用

qzwqbdag 于 2023-03-18 发布在其他

关注(0)|答案(2)|浏览(142)

我使用了“flow_from_directory”，但我的“lose”没有减少。我注意到当我运行“fit_generator”时，它显示有1个类，尽管我的掩码有3个类。我的问题是，我们是否需要在“datagen.flow_from_directory”中指示有多少个类？您是否看到“datagen.flow_from_directory”调用中的任何错误：

我的目录结构如下所示：

我的代码如下所示：

inputs = tf.keras.layers.Input(shape=(IMAGE_SIZE, IMAGE_SIZE, 3), name="input_image")

model  = tf.keras.applications.ResNet50(input_tensor=inputs, weights=None, include_top=true)

LR = 0.0001
optim = keras.optimizers.Adam(LR)

dice_loss_se2 = sm.losses.DiceLoss()
mae = tf.keras.losses.MeanAbsoluteError( )
metrics = [ mae,sm.metrics.IOUScore(threshold=0.5), sm.metrics.FScore(threshold=0.5) , dice_loss_se2]

model.compile(optimizer=optim,loss= dice_loss_se2,metrics= metrics)

image_datagen = ImageDataGenerator()
                
mask_datagen = ImageDataGenerator()
                 
image_generator =image_datagen.flow_from_directory( "/mydata/train/image", target_size=(IMAGE_SIZE, IMAGE_SIZE)
                                                   , class_mode = None,
                                                  )
                                                   

mask_generator = mask_datagen.flow_from_directory("/mydata/train/mask"  , target_size=(IMAGE_SIZE, IMAGE_SIZE)
                                                , class_mode = None,
                                                 )
                                                   

train_generator = zip(image_generator, mask_generator)

train_steps = 1212//batch_size

#---------------------------

image_generator_val =image_datagen.flow_from_directory( "/mydata/Validation/image", target_size=(IMAGE_SIZE, IMAGE_SIZE)
                                                   , class_mode = None,
                                                  )
                                                    

mask_generator_val = mask_datagen.flow_from_directory("/mydata/Validation/mask"  , target_size=(IMAGE_SIZE, IMAGE_SIZE)
                                                , class_mode = None,
                                                 )
                                                  )

val_generator = zip(image_generator_val, mask_generator_val)

val_steps = 250//batch_size


history =model.fit_generator(train_generator, validation_data=val_generator , steps_per_epoch=train_steps, validation_steps=val_steps , epochs=epochs, verbose=1)

keras

来源：https://stackoverflow.com/questions/71304293/using-datagen-flow-from-directory-with-image-segmination-and-number-of-classes

2条答案

按热度按时间

e5nqia271#

你的问题出在你的目录结构上。你需要的是如下所示的目录结构

mydata
---- train
     ---- image
          ------1.jpg
          ------2.jpg

     ---- mask
          ------1.png
          ------2.png

你只得到一个类，因为生成器只看到img类。2所以只要按照上面的目录结构移动你的图像

赞(0）回复(0）举报 2023-03-18

piwo6bdm2#

他们还做了一种方式，特定子集的训练或验证或指定文件夹，我的文件夹结构（目录）是见下.

F:\datasets\downloads\example\image
F:\datasets\downloads\example\image\Bee
F:\datasets\downloads\example\image\Shiny Jumbo
F:\datasets\downloads\example\image\Sleepy cat
...

def gen():
    train_generator = ImageDataGenerator(
            rescale=1./255,
            shear_range=0.2,
            zoom_range=0.2,
            horizontal_flip=True)
    train_generator = train_generator.flow_from_directory(
            directory,
            target_size=(150, 150),
            batch_size=32,
            class_mode='binary',    # None  # categorical   # binary
            subset='training')
    target = np.array([[i] for i in range(10)])
            
    return train_generator

train_generator = gen()
val_generator = train_generator

inputs = tf.keras.layers.Input(shape=(150, 150, 3), name="input_image")
model  = tf.keras.applications.ResNet50(input_tensor=inputs, weights=None, include_top=True)

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Optimizer
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
optimizer = tf.keras.optimizers.Nadam(
    learning_rate=0.0001, beta_1=0.9, beta_2=0.999, epsilon=1e-07,
    name='Nadam'
) # 0.00001

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Loss Fn
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""                               
# 1
# lossfn = tf.keras.losses.MeanSquaredLogarithmicError(reduction=tf.keras.losses.Reduction.AUTO, name='mean_squared_logarithmic_error')
# 2
lossfn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Model Summary
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
model.compile(optimizer=optimizer, loss=lossfn, metrics=['accuracy'])

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Training
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
history = model.fit_generator(train_generator, validation_data=val_generator, steps_per_epoch=train_steps, validation_steps=val_steps , epochs=epochs, verbose=1)

input('...')

找到10张图片，属于10个类。
Output example

赞(0）回复(0）举报 2023-03-18

我来回答

keras 将datagen.flow_from_directory与图像分割和类数结合使用

2条答案

相关问题

热门标签

最新问答