我想使用Massachusetts数据集进行训练，但是Massachusetts数据集中的图片格式是.tiff格式，我尝试自己修改了部分代码，但是依然报错，请问应该怎么修改相关部分？

Hi, there is a small correction in the unet code

In Dblock the nn.Conv2d accepts integer channel sizes so channel/2 is not acceptable, changing it to int(channel/2) works though

Traceback (most recent call last):
File "test.py", line 149, in
mask = solver.test_one_img_from_path(source+name)
File "test.py", line 32, in test_one_img_from_path
return self.test_one_img_from_path_2(path)
File "test.py", line 94, in test_one_img_from_path_2
img90 = np.array(np.rot90(img))
File "/home/yangzihao/.conda/envs/py27/lib/python2.7/site-packages/numpy/lib/function_base.py", line 121, in rot90
.format(axes, m.ndim))
ValueError: Axes=(0, 1) out of range for array of ndim=0.

I don't konw how to solve this, please help me orz

The label image will be read as NoneType when running this line:
mask = cv2.imread(os.path.join(root+'{}_mask.png').format(id), cv2.IMREAD_GRAYSCALE)
Maybe it should be changed like this:
mask = cv2.imread(os.path.join(root, '{}_mask.png').format(id), cv2.IMREAD_GRAYSCALE)

What changes are requited in the current script to use for multi class problem?

我改了之后发现不能用能教我一下吗

我在复现您的代码后，申请了对应的比赛但是一直没有收到批复，无法对模型输出的验证集的mask进行iou精度评估。
或者是否向您借一下codalab的账号以提交结果进行精度评估？
谢谢！

请问如何用两块GPU训练呢？？应该怎么改代码呢？

我能看看你的init里面的程序吗

Hello, there is a small thing I would like to ask you. Since the verification set provided by the DeepGlobal game does not have a GT image, I want to get the IOU of my own training model on the verification set. How should I submit the results of my prediction? I look forward to your reply and wish you a happy life.

I wanted to inquire under what license this code is issued?. As I do not see a LICENSE file and I would like to request that one is added. MIT or Apache would work for me, but of course please use whatever license terms you want

when i run python train.py using python virtual environment it shows the error. how to solve the error,anyone solved it? thanks a lot . Another, i have tried to use small batch size,but is doesn't work.

<torch.utils.data.dataloader.DataLoader object at 0x7f3868b7d790>
THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1512946747676/work/torch/lib/THC/generic/THCStorage.cu line=58 error=2 : out of memory
Traceback (most recent call last):
File "train.py", line 47, in
train_loss = solver.optimize()
File "/mnt/saners-extend/DeepGlobe-Road-Extraction-Challenge/framework.py", line 62, in optimize
pred = self.net.forward(self.img)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 66, in forward
return self.module(*inputs[0], **kwargs[0])
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/modules/module.py", line 325, in call
result = self.forward(*input, **kwargs)
File "/mnt/saners-extend/DeepGlobe-Road-Extraction-Challenge/networks/dinknet.py", line 176, in forward
e3 = self.encoder3(e2)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/modules/module.py", line 325, in call
result = self.forward(*input, **kwargs)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/modules/container.py", line 67, in forward
input = module(input)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/modules/module.py", line 325, in call
result = self.forward(*input, **kwargs)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torchvision-0.1.9-py2.7.egg/torchvision/models/resnet.py", line 41, in forward
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/modules/module.py", line 325, in call
result = self.forward(*input, **kwargs)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/modules/conv.py", line 277, in forward
self.padding, self.dilation, self.groups)
File "/home/anaconda3/envs/road_extract/lib/python2.7/site-packages/torch/nn/functional.py", line 90, in conv2d
return f(input, weight, bias)
RuntimeError: cuda runtime error (2) : out of memory at /opt/conda/conda-bld/pytorch_1512946747676/work/torch/lib/THC/generic/THCStorage.cu:58

谁有，能提供一下吗

请问您在训练过程中有没有遇到推理结构全白或者全黑的情形？这可能是什么原因导致的呢？分析了很多原因，做了很多修改和尝试，但推理结果依然是全白全黑，该怎么处理呢

Hello zlkanata,

I'm Yooseung Wang, ML Research Intern @ SIA Lab in South Korea.
We're inspired by your works on DeepGlobe Challenge what you've done so that we're working on DeepGlobe datasets to improve performances starting from your models, D-LinkNet. The reason why I mail you is that we want to compare our results with yours,especially in figure 3 of your D-LinkNet Workshop Paper.(CVPRW). Could you please give me the raw files of figure 3 (16 raw files, example results of three models.) in your D-LinkNet paper which had been published @ CVPRW. Of course Thanks a lot if you give me all outputs for three models (2334 images3(UNet,LinkNet34D-LinkNet34)).
Hope your great successes in your future work!
my email address is here : [email protected]

Pretty thanks for your awesome contribution!
But would it be possible for you to provide the weights that were already trained? Since that might save us a lot of training time and help us to evaluate the code?
I once tried to solver.load('weights/weights') in test.py but it failed 0.0

错误出现在train.py中低34行左右函数dataloader中data_loader = torch.utils.data.DataLoader(
dataset,
batch_size=batchsize,
shuffle=True,
num_workers=4,
)

ValueError: num_samples should be a positive integer value, but got num_samples=0
我的运行环境为：python3.6 pytorch1.3.1
打扰您了，期待您的回复，祝您在新的一年心想事成，平安喜乐~~

Hello, I have a question when running Python3 train.py, here is the error:

Traceback (most recent call last):
File "train.py", line 39, in
num_workers=4)
File "/home/ev1-ws4/anaconda3/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 176, in init
sampler = RandomSampler(dataset)
File "/home/ev1-ws4/anaconda3/lib/python3.5/site-packages/torch/utils/data/sampler.py", line 66, in init
"value, but got num_samples={}".format(self.num_samples))
ValueError: num_samples should be a positive integer value, but got num_samples=0

I output the length of "imagelist" and "trainlist", which is 6226 and 0, respectively.

imagelist = filter(lambda x: x.find('sat')!=-1, os.listdir(ROOT)) #length: 6226
trainlist = map(lambda x: x[:-8], imagelist) #length: 0
x = list(imagelist)
y = list(trainlist)
print(len(x), len(y))

Any ideas how to solve this problem? Many thanks.

Hi
I have trained the d-link model on the training images from DeepGlobe data. Now I'm trying to run the test on the data from Roadtracer images, which are 4096x4096. I changed the resolution in the test.py file. However, it eventually gives a CUDA memory error, despite having 10 GB of GPU memory. Do you have any suggestions to make this work?

当我运行trian.py的时候出现这个问题，请问北邮大神遇到过吗？需要如何解决？
Traceback (most recent call last):
File "ttest.py", line 71, in
data_loader_iter = iter(data_loader)
File "/home/Software/anaconda3/envs/python-3.5/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 310, in iter
return DataLoaderIter(self)
File "/home/Software/anaconda3/envs/python-3.5/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 180, in init
self._put_indices()
File "/home/Software/anaconda3/envs/python-3.5/lib/python3.5/site-packages/torch/utils/data/dataloader.py", line 219, in _put_indices
indices = next(self.sample_iter, None)
File "/home/Software/anaconda3/envs/python-3.5/lib/python3.5/site-packages/torch/utils/data/sampler.py", line 119, in iter
for idx in self.sampler:
File "/home/Software/anaconda3/envs/python-3.5/lib/python3.5/site-packages/torch/utils/data/sampler.py", line 50, in iter
return iter(torch.randperm(len(self.data_source)).long())
RuntimeError: invalid argument 1: must be strictly positive at /opt/conda/conda-bld/pytorch_1512954043090/work/torch/lib/TH/generic/THTensorMath.c:2184

https://www.baiten.cn/patent/detail/4e22fa2caab81ed9946867aa7fe32cb7bef5914376be05b9?sc=&fq=&type=&sort=&sortField=&q=dlinknet&rows=10#1/CN201910902885.7/detail/abst

Traceback (most recent call last):
File "C:\DeepGlobe-Road-Extraction-Challenge-master\train.py", line 31, in
data_loader = torch.utils.data.DataLoader(
File "C:\software\Python39\lib\site-packages\torch\utils\data\dataloader.py", line 266, in init
sampler = RandomSampler(dataset, generator=generator) # type: ignore
File "C:\software\Python39\lib\site-packages\torch\utils\data\sampler.py", line 103, in init
raise ValueError("num_samples should be a positive integer "
ValueError: num_samples should be a positive integer value, but got num_samples=0

so how could this be resolved?

请教一个问题，训练集图片这么大，训练时的batch_size应该是比较小的，那么BN层需要冻结吗，当时您是怎么做的

我在复现您的代码后，申请了对应的比赛但是一直没有收到批复，无法对模型输出的验证集的mask进行iou精度评估。
是否可以向您借一下codalab的账号以提交结果进行精度评估？
我的邮箱：[email protected]
谢谢！

① (when num_workers=4)
raise RuntimeError('DataLoader worker (pid(s) {}) exited unexpectedly'.format(pids_str)) from e
RuntimeError: DataLoader worker (pid(s) 12488) exited unexpectedly
② (when num_workers=0)
Process finished with exit code -1073741795 (0xC000001D)

anyone please who can help

$ python test.py
0 0.00
Traceback (most recent call last):
File "test.py", line 150, in
mask[mask>4.0] = 255
TypeError: 'NoneType' object does not support item assignment

这个项目没有添加val loss吗？矮瑞巴迪

mask[mask > XX] = 255, mask[mask <=XX] = 0, is the value of XX an experience value? Does the XX threshold need to be set differently for different graphs?

Hi, when I am running the train.py file it throws an error

AssertionError:
Found no NVIDIA driver on your system. Please check that you
have an NVIDIA GPU and installed a driver from
http://www.nvidia.com/Download/index.aspx

As I don't have GPU on my system, how can I run this code?

请问您可以提供官方数据集的验证集标签数据吗，我在复现您的代码后，申请了对应的比赛但是一直没有收到批复，无法对模型输出的验证集的mask进行iou精度评估。
或者是否向您借一下codalab的账号以提交结果进行精度评估？
谢谢！

Hi

Is the code for computing IoU and mIoU also included in the scripts, or does that have to be done separately from the resultant predictions in the submits directory?

@zlkanata 您好，作为一个初学者，我在使用你的代码的时候有几个疑问，不知道你能不能抽空回答一下：
1.代码里我没找到计算iou的部分，请问您能再上传一下吗。
2.我下载了官方的数据集，验证集没有标签，所以感到很困惑，怎么来验证模型的效果以及计算iou。
谢谢了~

Why the batch_size for test is 4. I see that you concatenate the image together. Can I use the batchsize equal 1?

我使用Massachusetts数据集进行训练的时候出现该报错，应该如何解决啊

我使用的是readme给模型，在运行test.py文件之后生成对于验证集的预测mask，发现好多mask都是空白的，并且其他的有黑色mask的结果并不是卫星图片当中的道路，这个是我代码改错了吗？我仅仅resize了一下图片的结果

def test_one_img_from_path_4(self, path):
    img = cv2.imread(path)#.transpose(2,0,1)[None]
    img=cv2.resize(img,(1024,1024),interpolation=cv2.INTER_LINEAR)
    img90 = np.array(np.rot90(img))
    img90=cv2.resize(img90,(1024,1024),interpolation=cv2.INTER_LINEAR)
    img1 = np.concatenate([img[None],img90[None]])

您好，有幸拜读了您的代码和论文，想请问您在归一化时 “ *3.2 - 1.6 “这一步是出于什么考虑呢？

Thanks for your outstanding work. I am getting weird errors during unet training would you please give me any suggestions?
File "/home/jhone/1_DeepUnet/pytorch_version/utils.py", line 36, in train
model = Dunet().to(device).float()
File "/home/jhone/1_DeepUnet/pytorch_version/model.py", line 49, in init
self.dilate_center = Dblock(512)
File "/home/jhone/1_DeepUnet/pytorch_version/model.py", line 15, in init
self.dilate1 = nn.Conv2d(channel/2, channel, kernel_size=3, dilation=1, padding=1)
File "/home/jhone/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 338, in init
False, _pair(0), groups, bias, padding_mode)
File "/home/jhone/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 48, in init
out_channels, in_channels // groups, *kernel_size))
TypeError: new(): argument 'size' must be tuple of ints, but found element of type float at pos 2
(torch) ali@ali-System

一个错解决了又出现另一个，无穷无尽

您好，UNet多次修改学习率（e-3/-4/-5/-6/-7) 训练，但是预测结果图全黑，请问 该怎么办呢？ 按照网上说的方法修改后，还是全黑。只有lr=e-6，epoch=200时预测结果显示，但是效果非常不好！

Hi, Thank you for your wonderful work! May I ask what's your logic for the 'test_one_img_from_path_1' method in test.py? Specifically, why did you need to do the following before calling net.forward?

img90 = np.array(np.rot90(img))
img1 = np.concatenate([img[None],img90[None]])
img2 = np.array(img1)[:,::-1]
img3 = np.concatenate([img1,img2])
img4 = np.array(img3)[:,:,::-1]
img5 = np.concatenate([img3,img4]).transpose(0,3,1,2)
img5 = np.array(img5, np.float32)/255.0 * 3.2 -1.6

DLink原文是8，看到代码里加了一个16的版本，是否有做过相应的精度对比实验呢？

Hi, thanks for your inspiring work! Just one point I don't understand; why do you set
img = np.array(img, np.float32).transpose(2,0,1)/255.0 * 3.2 - 1.6
when training? Is there any advantages for this operation?