训练的时候并没有出什么问题，但是在训练过程结束之后出现了下面这些提示：
Epoch 30/30
703/703 [==============================] - 1410s 2s/step - loss: 0.2347 - acc: 0.9036 - val_loss: 1.0093 - val_acc: 0.7139
Traceback (most recent call last):

File "", line 1, in
get_ipython().run_line_magic('run', 'segnet_train.py -m segnet.h5')

File "/opt/Anaconda3/lib/python3.7/site-packages/IPython/core/interactiveshell.py", line 2287, in run_line_magic
result = fn(*args,**kwargs)

File "", line 2, in run

File "/opt/Anaconda3/lib/python3.7/site-packages/IPython/core/magic.py", line 187, in
call = lambda f, *a, **k: f(*a, **k)

File "/opt/Anaconda3/lib/python3.7/site-packages/IPython/core/magics/execution.py", line 807, in run
run()

File "/opt/Anaconda3/lib/python3.7/site-packages/IPython/core/magics/execution.py", line 793, in run
exit_ignore=exit_ignore)

File "/opt/Anaconda3/lib/python3.7/site-packages/IPython/core/pylabtools.py", line 180, in mpl_execfile
if plt.draw_if_interactive.called:

AttributeError: 'function' object has no attribute 'called'

我真的是看不出来是哪里的问题……有人知道吗？

如果一次训练一个类别，那么label数据可是包含5个类别，是不是需要处理一下呢？

unet_train.py我改动了一下

`classes_dict={1:'VEGETATION',2:'BUILDING',3:'WATER',4:'ROAD'}

def generateData(batch_size,classtype,data=[]):
while True:
train_data = []
train_label = []
batch = 0
for i in (range(len(data))):
url = data[i]
batch += 1
img = load_img(filepath + 'src/' + url)
img = img_to_array(img)
train_data.append(img)
label = load_img(filepath + 'label/' + url, grayscale=True)
################################################################
for key in classes_dict.keys():
if(key != classtype):
mask=(label==key)
label[mask]=0
################################################################
label = img_to_array(label)
train_label.append(label)
if batch % batch_size==0:
train_data = np.array(train_data)
train_label = np.array(train_label)
yield (train_data,train_label)
train_data = []
train_label = []
batch = 0

def generateValidData(batch_size,classtype,data=[]):
while True:
valid_data = []
valid_label = []
batch = 0
for i in (range(len(data))):
url = data[i]
batch += 1
img = load_img(filepath + 'src/' + url)
img = img_to_array(img)
valid_data.append(img)
label = load_img(filepath + 'label/' + url, grayscale=True)
################################################################
for key in classes_dict.keys():
if(key != classtype):
mask=(label==key)
label[mask]=0
################################################################
label = img_to_array(label)
valid_label.append(label)
if batch % batch_size==0:
valid_data = np.array(valid_data)
valid_label = np.array(valid_label)
yield (valid_data,valid_label)
valid_data = []
valid_label = []
batch = 0
`

ValueError: Negative dimension size caused by subtracting 2 from 1 for 'max_pooling2d_2/MaxPool' (op: 'MaxPool') with input shapes: [?,1,128,128].
楼主该怎么处理？？

用自己的数据训练unet，结果一直很差且保持不变，这是什么原因？

训练U-Net的时候似乎参数有问题，网络应该没有在学习，请问大家遇到了同样的情况么？怎么解决的？
703/703 [==============================] - 449s - loss: -4.4077 - acc: 0.3267 - val_loss: -4.3856 - val_acc: 0.3299
Epoch 2/30
703/703 [==============================] - 444s - loss: -4.4410 - acc: 0.3268 - val_loss: -4.3856 - val_acc: 0.3299
Epoch 3/30
703/703 [==============================] - 445s - loss: -4.4410 - acc: 0.3268 - val_loss: -4.3856 - val_acc: 0.3299
Epoch 4/30
703/703 [==============================] - 444s - loss: -4.4410 - acc: 0.3268 - val_loss: -4.3856 - val_acc: 0.3299

迭代了30次，数值未曾变化过

When I changed my own data set and executed the training code segnet_train.py, I encountered the following error:
Traceback (most recent call last):
File "segnet_train.py", line 284, in
train(args)
File "segnet_train.py", line 247, in train
validation_data=generateValidData(BS,val_set), steps_per_epoch=train_numb//BS, max_queue_size=1)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/keras/legacy/interfaces.py", line 91, in wrapper
return func(*args, **kwargs)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/keras/engine/training.py", line 1418, in fit_generator
initial_epoch=initial_epoch)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/keras/engine/training_generator.py", line 181, in fit_generator
generator_output = next(output_generator)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/keras/utils/data_utils.py", line 709, in get
six.reraise(*sys.exc_info())
File "/home/gnss/anaconda3/lib/python3.5/site-packages/six.py", line 693, in reraise
raise value
File "/home/gnss/anaconda3/lib/python3.5/site-packages/keras/utils/data_utils.py", line 685, in get
inputs = self.queue.get(block=True).get()
File "/home/gnss/anaconda3/lib/python3.5/multiprocessing/pool.py", line 608, in get
raise self._value
File "/home/gnss/anaconda3/lib/python3.5/multiprocessing/pool.py", line 119, in worker
result = (True, func(*args, **kwds))
File "/home/gnss/anaconda3/lib/python3.5/site-packages/keras/utils/data_utils.py", line 626, in next_sample
return six.next(_SHARED_SEQUENCES[uid])
File "segnet_train.py", line 108, in generateData
train_label = labelencoder.transform(train_label)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/sklearn/preprocessing/label.py", line 257, in transform
_, y = encode(y, uniques=self.classes, encode=True)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/sklearn/preprocessing/label.py", line 110, in _encode
return _encode_numpy(values, uniques, encode)
File "/home/gnss/anaconda3/lib/python3.5/site-packages/sklearn/preprocessing/label.py", line 53, in _encode_numpy
% str(diff))
ValueError: y contains previously unseen labels: [7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 15.0]
This seems to be a problem with the label? Thank you for helping me.

Will the 1.png, 2.png to 5.png refer to the original 5 remote sensing images in the gen_dataset.py? But there are no five images in the downloaded data. What's the matter?Thank you for your reply

你好，请问按照你的步骤训练segnet没有什么改动，但segnet训练后验证集loss在0.8左右一直下不来是为什么呢？

咦，为什么我的结果全是0呢，你有没有遇到这种情况呀，或者你改动了程序的什么地方呢。我都被困扰了好多天了
…
------------------ 原始邮件 ------------------ 发件人: "GusRoth"[email protected]; 发送时间: 2019年3月6日(星期三) 下午5:25 收件人: "AstarLight/Satellite-Segmentation"[email protected]; 抄送: "cup_zyx"[email protected]; "Comment"[email protected]; 主题: Re: [AstarLight/Satellite-Segmentation] use unet_predict.py formulti-classification (#27) @GusRoth 你好，请问你用U_net跑出结果了吗？ 跑出来了，不过结果看上去并没有作者的那么好，有很大的误差，不知道哪里出了问题 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

你是用的作者的数据集么？我用作者的数据集出现了我上边描述的结果，没有做什么更改，只是改掉了我上面提问的那个错误，就是一个channel_first的问题。就通了。
然后你说的全零的情况我遇到过，我跑自己的数据的时候就是这样子。
开始是因为我要预测的图给错了，后来换成正确的，还处理了一下数据集，虽然结果很差，但是没有全零的情况了。

Originally posted by @GusRoth in #27 (comment)

Excuse me, what should the sequence of programs run, I have made a mistake now:
usage: unet_train.py [-h] [-d DATA] -m MODEL [-p PLOT]
Unet_train.py: error: the following arguments are required: -m/--model,
However, the models under my .keras contain the downloaded weight model, and I also redefined the address in the parameters of ModelCheckpoint(), but I still get an error, why, I hope to get your reply, thank you

大家，我在复现代码时，发现一个问题就是，误检率非常之高，基本什么都能被识别为建筑物，请问该如何解决这个问题

我看两个文件里生成图片的代码都一样，unet训练不能用segnet的吗

@james Lee, hi，非常感谢你代码，学习了很多。经历了一点小挫折，最终还是跑通代码。作为简单的实验，我用gen_dataset.py 生成了2000个训练数据对，，然后精简了u-net的网络
Layer (type) Output Shape Param # Connected to

input_2 (InputLayer) (None, 3, 256, 256) 0

conv2d_12 (Conv2D) (None, 32, 256, 256) 896 input_2[0][0]

max_pooling2d_3 (MaxPooling2D) (None, 32, 128, 128) 0 conv2d_12[0][0]

conv2d_13 (Conv2D) (None, 64, 128, 128) 18496 max_pooling2d_3[0][0]

max_pooling2d_4 (MaxPooling2D) (None, 64, 64, 64) 0 conv2d_13[0][0]

conv2d_14 (Conv2D) (None, 128, 64, 64) 73856 max_pooling2d_4[0][0]

up_sampling2d_3 (UpSampling2D) (None, 128, 128, 128 0 conv2d_14[0][0]

concatenate_3 (Concatenate) (None, 192, 128, 128 0 up_sampling2d_3[0][0]
conv2d_13[0][0]

conv2d_15 (Conv2D) (None, 64, 128, 128) 110656 concatenate_3[0][0]

up_sampling2d_4 (UpSampling2D) (None, 64, 256, 256) 0 conv2d_15[0][0]

concatenate_4 (Concatenate) (None, 96, 256, 256) 0 up_sampling2d_4[0][0]
conv2d_12[0][0]

conv2d_16 (Conv2D) (None, 32, 256, 256) 27680 concatenate_4[0][0]

conv2d_17 (Conv2D) (None, 1, 256, 256) 33 conv2d_16[0][0]

Total params: 231,617
Trainable params: 231,617
Non-trainable params: 0
然后发现数据训练val_acc: 0.4705最高了，然后多次epoch都是0.4705没有进一步提高。
同时测试的结果mask输出都是1。感觉网络好像没有学习到有效的分割信息。
所以想问一下这个情况，会有哪些可能性？可以如何解决？

你好，在我巡行segnet_train的时候出现没有回调函数这个对象的错误，模型能够保存出来，但是无法绘制出精度和损失率的图像，请问该怎样解决

I meet some errors like this:
ValueError: Negative dimension size caused by subtracting 2 from 1 for 'max_pooling2d_2/MaxPool' (op: 'MaxPool') with input shapes: [?,1,128,64].
what's the reason behind it ,how to solve it? Thank you!

Hello, thank you very much for sharing the codes. I want to know the parameter setting for the segnet and unet. The parameters are same with that in your codes? Or may be you can share the trained model to me. Waiting for your reply, thank you again. Good Luck!

我试了两种方法，第一种用labelme对我自己的原始两张遥感大图做标签，转化为mask图像后像素值范围是0-255，故利用划分像素值的方法把图像分为0、1两种像素（我的目前只有一类），预测出来的结果把1标签对应255像素转化过来，0像素不变，图像显示很乱，多是锯齿状，没有正常图片的圆滑，不知道是我训练的过程有问题还是标签有问题还是我最后预测结果可视化的问题，现在想请教一下大神们标签是怎么做的？
（第二种方法用labelme做完标签之后利用label.png转化为0、1像素，昨晚试验了一下，貌似这种方法训练不通）

Hello
I have a question want to ask for you
If I define a new scalars, how can I visualize it on Tensorboard in Keras?
such as I define a new variable learning_rate, It changes over epoch, how I can visualize it on tensorboard?
I used keras, backend is tensorflow.
Looking forward to your reply.

Traceback (most recent call last):
File "segnet_train.py", line 256, in
train(args)
File "segnet_train.py", line 212, in train
model = SegNet()
File "segnet_train.py", line 200, in SegNet
model.add(Reshape((n_label,img_w*img_h)))
File "/home/gumingqi/anaconda2/envs/segnet/lib/python2.7/site-packages/keras/models.py", line 522, in add
output_tensor = layer(self.outputs[0])
File "/home/gumingqi/anaconda2/envs/segnet/lib/python2.7/site-packages/keras/engine/topology.py", line 638, in call
output_shape = self.compute_output_shape(input_shape)
File "/home/gumingqi/anaconda2/envs/segnet/lib/python2.7/site-packages/keras/layers/core.py", line 403, in compute_output_shape
input_shape[1:], self.target_shape)
File "/home/gumingqi/anaconda2/envs/segnet/lib/python2.7/site-packages/keras/layers/core.py", line 391, in _fix_unknown_dimension
raise ValueError(msg)
ValueError: total size of new array must be unchanged
another question is if the backend of keras is theano?
thank u!

please help me.
I don't konw how to solve this, why there is a 3 in width

Thank you for your code, I learned a lot from it !
I visited your blog , it is said that there are 5 remote sensing images for training, but I only got 2 images from your Baidu Netdisk. Could you share all images? Thank you !

不应该是 img_h , img_w, 3 吗？
这样也更符合图像在内存的分布吧？
predict会出现错误

Could someone help ?

Error:
ValueError: Error when checking target: expected conv2d_19 to have 4 dimensions, but got array with shape (5, 256, 256)

Summary :
Layer (type) Output Shape Param # Connected to
input_1 (InputLayer) (None, 256, 256, 3) 0
conv2d_1 (Conv2D) (None, 256, 256, 32) 896 input_1[0][0]
conv2d_2 (Conv2D) (None, 256, 256, 32) 9248 conv2d_1[0][0]
max_pooling2d_1 (MaxPooling2D) (None, 128, 128, 32) 0 conv2d_2[0][0]
conv2d_3 (Conv2D) (None, 128, 128, 64) 18496 max_pooling2d_1[0][0]
conv2d_4 (Conv2D) (None, 128, 128, 64) 36928 conv2d_3[0][0]
max_pooling2d_2 (MaxPooling2D) (None, 64, 64, 64) 0 conv2d_4[0][0]
conv2d_5 (Conv2D) (None, 64, 64, 128) 73856 max_pooling2d_2[0][0]
conv2d_6 (Conv2D) (None, 64, 64, 128) 147584 conv2d_5[0][0]
max_pooling2d_3 (MaxPooling2D) (None, 32, 32, 128) 0 conv2d_6[0][0]
conv2d_7 (Conv2D) (None, 32, 32, 256) 295168 max_pooling2d_3[0][0]
conv2d_8 (Conv2D) (None, 32, 32, 256) 590080 conv2d_7[0][0]
max_pooling2d_4 (MaxPooling2D) (None, 16, 16, 256) 0 conv2d_8[0][0]
conv2d_9 (Conv2D) (None, 16, 16, 512) 1180160 max_pooling2d_4[0][0]
conv2d_10 (Conv2D) (None, 16, 16, 512) 2359808 conv2d_9[0][0]
up_sampling2d_1 (UpSampling2D) (None, 32, 32, 512) 0 conv2d_10[0][0]
concatenate_1 (Concatenate) (None, 32, 32, 768) 0 up_sampling2d_1[0][0]
conv2d_8[0][0]
conv2d_11 (Conv2D) (None, 32, 32, 256) 1769728 concatenate_1[0][0]
conv2d_12 (Conv2D) (None, 32, 32, 256) 590080 conv2d_11[0][0]
up_sampling2d_2 (UpSampling2D) (None, 64, 64, 256) 0 conv2d_12[0][0]
concatenate_2 (Concatenate) (None, 64, 64, 384) 0 up_sampling2d_2[0][0]
conv2d_6[0][0]
conv2d_13 (Conv2D) (None, 64, 64, 128) 442496 concatenate_2[0][0]
conv2d_14 (Conv2D) (None, 64, 64, 128) 147584 conv2d_13[0][0]
up_sampling2d_3 (UpSampling2D) (None, 128, 128, 128 0 conv2d_14[0][0]
concatenate_3 (Concatenate) (None, 128, 128, 192 0 up_sampling2d_3[0][0]
conv2d_4[0][0]
conv2d_15 (Conv2D) (None, 128, 128, 64) 110656 concatenate_3[0][0]
conv2d_16 (Conv2D) (None, 128, 128, 64) 36928 conv2d_15[0][0]
up_sampling2d_4 (UpSampling2D) (None, 256, 256, 64) 0 conv2d_16[0][0]
concatenate_4 (Concatenate) (None, 256, 256, 96) 0 up_sampling2d_4[0][0]
conv2d_2[0][0]
conv2d_17 (Conv2D) (None, 256, 256, 32) 27680 concatenate_4[0][0]
conv2d_18 (Conv2D) (None, 256, 256, 32) 9248 conv2d_17[0][0]
conv2d_19 (Conv2D) (None, 256, 256, 1) 33 conv2d_18[0][0]

Total params: 7,846,657
Trainable params: 7,846,657
Non-trainable params: 0

Code:
`

def unet():
inputs = Input((img_w, img_h,3))

conv1 = Conv2D(32, (3, 3), activation="relu", padding="same")(inputs)
conv1 = Conv2D(32, (3, 3), activation="relu", padding="same")(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)

conv2 = Conv2D(64, (3, 3), activation="relu", padding="same")(pool1)
conv2 = Conv2D(64, (3, 3), activation="relu", padding="same")(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)

conv3 = Conv2D(128, (3, 3), activation="relu", padding="same")(pool2)
conv3 = Conv2D(128, (3, 3), activation="relu", padding="same")(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)

conv4 = Conv2D(256, (3, 3), activation="relu", padding="same")(pool3)
conv4 = Conv2D(256, (3, 3), activation="relu", padding="same")(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)

conv5 = Conv2D(512, (3, 3), activation="relu", padding="same")(pool4)
conv5 = Conv2D(512, (3, 3), activation="relu", padding="same")(conv5)

up6 = concatenate([UpSampling2D(size=(2, 2))(conv5), conv4], axis=3)
conv6 = Conv2D(256, (3, 3), activation="relu", padding="same")(up6)
conv6 = Conv2D(256, (3, 3), activation="relu", padding="same")(conv6)

up7 = concatenate([UpSampling2D(size=(2, 2))(conv6), conv3], axis=3)
conv7 = Conv2D(128, (3, 3), activation="relu", padding="same")(up7)
conv7 = Conv2D(128, (3, 3), activation="relu", padding="same")(conv7)

up8 = concatenate([UpSampling2D(size=(2, 2))(conv7), conv2], axis=3)
conv8 = Conv2D(64, (3, 3), activation="relu", padding="same")(up8)
conv8 = Conv2D(64, (3, 3), activation="relu", padding="same")(conv8)

up9 = concatenate([UpSampling2D(size=(2, 2))(conv8), conv1], axis=3)
conv9 = Conv2D(32, (3, 3), activation="relu", padding="same")(up9)
conv9 = Conv2D(32, (3, 3), activation="relu", padding="same")(conv9)

conv10 = Conv2D(n_label, (1, 1), activation="sigmoid")(conv9)
#conv10= Conv2D(n_label, (1, 1), activation="softmax")(conv9)

model = Model(inputs=inputs, outputs=conv10)
model.compile(optimizer='Adam', loss='binary_crossentropy', metrics=['accuracy'])
return model

`

hi, I am dealing with change detection research this days. And i noticed that the changedetetion competition on tianchi,"https://tianchi.aliyun.com/competition/entrance/231615/information". But i can not download the dataset. Would you kindly provide the data of 2015y. Thanks a lot.~~

Hello, I now run the program will report such an error：
ValueError:Concatenate layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 32, 32, 512), (None, 32, 32, 256)].
How should this be solved? Looking forward to your reply, thank you

pred = model.predict(crop,verbose=0)
pred = pred.reshape((256,256)).astype(np.uint8)
使用unet_predict.py进行多分类预测时，到pred = pred.reshape((256,256)).astype(np.uint8)报错，意思是不能将256x256x5个字节转换为256x256,所以这里要改成pred = pred.reshape((256,256, 5)).astype(np.uint8)吗？

Thank you for your code, I learned a lot from it !
I visited your blog , it is useful to me that you process big image to 252*256 small pictures.
Would you share code of preprocess data?
Thanks.

将所有的input_shape改成 input_shape=(img_w,img_h,3) keras有input_shape 通道参数在前，和通道参数在后之分

运行gen_dataset.py后报错如下，此时设定的文件夹里文件数量不一定，每次运行完都清空，发现有时能出现1张，有时4张，或者9张。我是在windows python3.6下运行的
File "D:/untitled/nlp/gen_dataset.py", line 18, in gamma_transform
return cv2.LUT(img, gamma_table)
TypeError: Expected cv::UMat for argument 'src'
再次运行报错也不一定相同，如
File "D:/untitled/nlp/gen_dataset.py", line 18, in gamma_transform
return cv2.LUT(img, gamma_table)
TypeError: Expected cv::UMat for argument 'img'
另一次
temp_x = np.random.randint(0, img.shape[0])
AttributeError: 'tuple' object has no attribute 'shape'

还是有其他的方法先对原始数据进行处理？

我的也是这个问题 更换数据集之后loss一直是负的。@duan147 你好，我也在用这个U_net，程序能运行，但是我试了好多天结果一直全是0，请问你解决这个问题了吗？

Originally posted by @Jessaminzyx in #5 (comment)

我按照模型上跑，predict的照片全黑，是什么原因吗？

博主您好，我在跑segnet_predict.py的时候 出现invalid size的输出，请问这个怎么解决，同时会输出图片是全黑的，静候您的回答

请问楼主有训练好的模型吗？
实验室只有一个1050ti，我想看一下效果，自己训练不出来，希望楼主能上传一下训练好的模型

预处理后的数据集没有下载链接

Why need (2,256,256) ????????

在我将n_label设置为2，classe设置为【0.，1.】的时候进行二分类训练，出现了下面这个错误：ValueError:y contains previously unsee labeels: 38.0
我该如何解决它？麻烦大家帮帮忙，谢谢！

raise ValueError('Unsupported image shape: ', x.shape)
ValueError: ('Unsupported image shape: ', ())
I have this problem, can someone help me solve it? Thank you.

可以分享原始数据吗?

您好，非常感谢您把算法代码提供，我想在你代码基础上测试一下用Unet做多分类，但是在修改网络和generateTrainData时，总是不对，能否提供一下帮助，将Unet做多分类的代码提供，谢谢

when i was run the segnet.py raised this error:
raise ValueError('Unsupported image shape: ', x.shape)
ValueError: ('Unsupported image shape: ', ())
i would be grateful if you could answer this questions @AstarLight
thank you so much !

跑模型的时候CPU 100%，然后就停止了，想问一下怎么降低配置要求啊？或者最低配置大概是什么样啊，谢谢