我想不出解决这个图像分割问题的办法

Question

我的训练图像是由一些拼接图像的ELAs（Error Level Analysis）中提取的蓝色通道组成的，标签只是它们对应的地面真值掩码。

我已经构造了一个简单的编码器-解码器CNN下面给出做分割，并在细胞膜分割任务上测试了它。在那里它的表现很好，并且创建了接近地面的真实图像，所以我猜我创建的神经网络是足够的能力。

但是，它不能在CASIA1+CASIA1GroundTruth数据集上的拼接图像上工作。请帮助我修复它，我已经花了太多天在它上尝试不同的架构和预处理的图像，但没有运气。

首先，它声称高精度（98%）和低损耗，但输出的图像是如此错误。如果你仔细看的话，这就像是得到了想要的面具，但是伴随着它的还有很多白色飞溅的区域。似乎它无法获得所需区域的像素强度与背景的差异。请帮我修一下：（

def process(img):
    img=img.getchannel('B')
    return img

for i in splicedIMG:
  img=process(Image.open('ELAs/'+str(i)))
  X.append(np.array(img)/np.max(img))

for i in splicedGT:
  lbl=Image.open('SGTResized/'+str(i))
  Y.append(np.array(lbl)/np.max(lbl))

X = np.array(X)
Y = np.array(Y)

X = X.reshape(-1, 256,256, 1)
Y = Y.reshape(-1, 256,256, 1)

X_train, X_val, Y_train, Y_val = train_test_split(X, Y, test_size = 0.2)

model = Sequential()

model.add(Conv2D(filters = 16, kernel_size = (3,3),padding = 'same', 
                 activation ='relu', input_shape = (256,256,1)))
model.add(BatchNormalization())
model.add(Conv2D(filters = 16, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))

model.add(Conv2D(filters = 32, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(Conv2D(filters = 32, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))

model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))

model.add(Conv2D(filters = 128, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(Conv2D(filters = 128, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2,2)))

model.add(UpSampling2D(size = (2,2)))
model.add(Conv2D(filters = 128, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(Conv2D(filters = 128, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())

model.add(UpSampling2D(size = (2,2)))
model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())

model.add(UpSampling2D(size = (2,2)))
model.add(Conv2D(filters = 32, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())
model.add(Conv2D(filters = 32, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())

model.add(UpSampling2D(size = (2,2)))
model.add(Conv2D(filters = 16, kernel_size = (3,3),padding = 'same', 
                 activation ='relu', input_shape = (256,256,3)))
model.add(BatchNormalization())
model.add(Conv2D(filters = 16, kernel_size = (3,3),padding = 'same', 
                 activation ='relu'))
model.add(BatchNormalization())

model.add(Conv2D(filters = 1, kernel_size = (1,1), activation = 'sigmoid'))

model.summary()

model.compile(optimizer = Adam(lr = 0.0001), loss = 'binary_crossentropy', metrics = ['accuracy'])

model_checkpoint = ModelCheckpoint('segmenter_weights.h5', monitor='loss',verbose=1, save_best_only=True)

model.fit(X_train, Y_train, validation_data = (X_val, Y_val), batch_size=4, epochs=200, verbose=1, callbacks=[PlotLossesCallback(),model_checkpoint])

Answer 1

哎呀，我做了一个傻事。为了查看我从X数组中提取的测试内容，我将该数组乘以255，因为PIL不显示0-1范围内的数组。错误的是，我只是使用了相同的修改变量，并在测试/预测中通过了它。