3DUNet implemented with pytorch
File "/opt/data/private/3DUNet-Pytorch-master/dataset/transforms.py", line 153, in call
img, mask = t(img, mask)
File "/opt/data/private/3DUNet-Pytorch-master/dataset/transforms.py", line 64, in call
tmp_img[:,:es-ss] = img[:,ss:es]
RuntimeError: The expanded size of the tensor (48) must match the existing size (36) at non-singleton dimension 1. Target sizes: [1, 48, 256, 256]. Tensor sizes: [36, 256, 256]
正常读入数据集，出现这样的报错是什么原因呀？
您好~如题，我注意到Unet.py里似乎没有涉及到任何比如batch或者instance等等的normalization操作，但我记得原始的3D UNet里应该是有的。请问您这样做是有什么用意吗？还是说encoder某一步里有自带normalization呢？
盼复，十分感谢！
您好，最近在用您分享的代码做分割，train的背景dice0=0.9984,目标dice1= 0.6014，但是test的时候dice0=0.6787,dice1=0.0011，且预测出来的标签全黑。我是用4个GPU进行训练和预测的，请问出现这种问题的原因是什么呢？和模型参数的读取有关系？希望得到您的帮助！
Hello,I want to run you code on my computer,and I followed the instruction in README.md.
in the preprocess_LiTS.py file,line of 75 as below:
print('Too little slice，give up the sample:', ct_file)
there was an error like this:
Unresolved reference 'ct_file'
how can i solve the problem?
waiting for your rely
thanks
你好，我用你的网盘数据可以训练测试，但是用我自己的CT数据、分割数据，到这一步就卡住报错了：
fixd_path = r'D:\Download\3DUNet-Pytorch-master\fixed_data'
dataset = Lits_DataSet([16, 64, 64],0.5,fixd_path,mode='train') #batch size
data_loader=DataLoader(dataset=dataset,batch_size=2,num_workers=1, shuffle=True)
for batch_idx, (data, target) in enumerate(data_loader):
print("最大值：")
print(torch.max(target.long() , 0))
print(data.shape, target.shape)
target = to_one_hot_3d(target.long())
到to_one_hot_3d()函数里的scatter时报错
torch.Size([2, 1, 16, 64, 64]) torch.Size([2, 16, 64, 64])
Traceback (most recent call last):
File "D:/Download/3DUNet-Pytorch-master/dataset/dataset_lits.py", line 65, in
main()
File "D:/Download/3DUNet-Pytorch-master/dataset/dataset_lits.py", line 57, in main
target = to_one_hot_3d(target.long())
File "D:\Download\3DUNet-Pytorch-master\utils\common.py", line 28, in to_one_hot_3d
one_hot = torch.zeros(n, n_classes, s, h, w).scatter_(1, tensor.view(n, 1, s, h, w), 1)
RuntimeError: Invalid index in scatter at ..\aten\src\TH/generic/THTensorEvenMoreMath.cpp:551
Hello, I ran your code and appeared“RuntimeError: The size of tensor a (512) must match the size of tensor b (256) at non-singleton dimension 3”，Can you help me solve it？
RuntimeError: CUDA out of memory. Tried to allocate 38.81 GiB (GPU 0; 48.00 GiB total capacity; 21.21 GiB already allocated; 23.81 GiB free; 22.30 GiB reserved in total by PyTorch)
Why can't I run this code with 48GB of video memory?
CPU is a 48GB NVIDIA Quadro RTX 8000
你好，为什么在读取数据的时候，无法读取？报下面的错：
Traceback (most recent call last):
File "E:/PytorchProject/3DUNet-Pytorch-master/dataset/dataset_lits_train.py", line 67, in
for i, (ct, seg) in enumerate(train_dl):
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch\utils\data\dataloader.py", line 521, in next
data = self._next_data()
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch\utils\data\dataloader.py", line 1203, in _next_data
return self._process_data(data)
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch\utils\data\dataloader.py", line 1229, in _process_data
data.reraise()
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch_utils.py", line 434, in reraise
raise exception
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch\utils\data_utils\worker.py", line 287, in _worker_loop
data = fetcher.fetch(index)
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch\utils\data_utils\fetch.py", line 49, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\torch\utils\data_utils\fetch.py", line 49, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "E:\PytorchProject\3DUNet-Pytorch-master\dataset\dataset_lits_train.py", line 29, in getitem
ct = sitk.ReadImage(self.filename_list[index][0], sitk.sitkInt16)
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\SimpleITK\extra.py", line 346, in ReadImage
return reader.Execute()
File "D:\Anaconda3\envs\pytorch_gpu\lib\site-packages\SimpleITK\SimpleITK.py", line 8015, in Execute
return _SimpleITK.ImageFileReader_Execute(self)
RuntimeError: Exception thrown in SimpleITK ImageFileReader_Execute: D:\a\1\sitk\Code\IO\src\sitkImageReaderBase.cxx:97:
sitk::ERROR: The file "E:/Pytorch" does not exist.
dataset文件夹下的dataset_lists_train文件中对数据的transform操作怎么能保证对原图和分割图做的操作是一样的呢？
感觉这样transform是不对的，我改成别的方法了
in dataset_lits_train.py, def get_np_data_3d(),line44:
def get_np_data_3d(self, filename, resize_scale=1):
data_np = sitk.GetArrayFromImage(sitk.ReadImage(self.dataset_path + '/data/' + filename, sitk.sitkInt16))
if self.resize_scale !=1.0:
data_np = ndimage.zoom(data_np, zoom=self.resize_scale, order=3)
data_np = data_np.astype(np.float32) # --------不理解的地方二
data_np=data_np/200.0 #--------------不理解的地方一
i cannot understand the function of this code:'data_np=data_np/200.0'，why 200 divided by 3, how do we get the parameter 200? assign at random?or other reason?
Could anyone answer my question?
Hi lee-zp
Thanks for sharing your code.When I was training my own data, I had the following problem.My data is the data of MRI rectum, and I want to see the dice of tumor.
I tried some online methods, but nothing worked.Can you help me ?
Best wish
Thank you very much for the code you provided. Could you please tell me how to solve this problem in my mailbox "test"
Traceback (most recent call last):
File "test.py", line 52, in
model.load_state_dict(ckpt['net'])
File "/home/ubuntu/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1406, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for DataParallel:
Missing key(s) in state_dict: "module.encoder_stage1.0.weight", "module.encoder_stage1.0.bias", "module.encoder_stage1.1.weight", "module.encoder_stage1.2.weight", "module.encoder_stage1.2.bias", "module.encoder_stage1.3.weight", "module.encoder_stage2.0.weight", "module.encoder_stage2.0.bias", "module.encoder_stage2.1.weight", "module.encoder_stage2.2.weight", "module.encoder_stage2.2.bias", "module.encoder_stage2.3.weight", "module.encoder_stage2.4.weight", "module.encoder_stage2.4.bias", "module.encoder_stage2.5.weight", "module.encoder_stage3.0.weight", "module.encoder_stage3.0.bias", "module.encoder_stage3.1.weight", "module.encoder_stage3.2.weight", "module.encoder_stage3.2.bias", "module.encoder_stage3.3.weight", "module.encoder_stage3.4.weight", "module.encoder_stage3.4.bias", "module.encoder_stage3.5.weight", "module.encoder_stage4.0.weight", "module.encoder_stage4.0.bias", "module.encoder_stage4.1.weight", "module.encoder_stage4.2.weight", "module.encoder_stage4.2.bias", "module.encoder_stage4.3.weight", "module.encoder_stage4.4.weight", "module.encoder_stage4.4.bias", "module.encoder_stage4.5.weight", "module.decoder_stage1.0.weight", "module.decoder_stage1.0.bias", "module.decoder_stage1.1.weight", "module.decoder_stage1.2.weight", "module.decoder_stage1.2.bias", "module.decoder_stage1.3.weight", "module.decoder_stage1.4.weight", "module.decoder_stage1.4.bias", "module.decoder_stage1.5.weight", "module.decoder_stage2.0.weight", "module.decoder_stage2.0.bias", "module.decoder_stage2.1.weight", "module.decoder_stage2.2.weight", "module.decoder_stage2.2.bias", "module.decoder_stage2.3.weight", "module.decoder_stage2.4.weight", "module.decoder_stage2.4.bias", "module.decoder_stage2.5.weight", "module.decoder_stage3.0.weight", "module.decoder_stage3.0.bias", "module.decoder_stage3.1.weight", "module.decoder_stage3.2.weight", "module.decoder_stage3.2.bias", "module.decoder_stage3.3.weight", "module.decoder_stage3.4.weight", "module.decoder_stage3.4.bias", "module.decoder_stage3.5.weight", "module.decoder_stage4.0.weight", "module.decoder_stage4.0.bias", "module.decoder_stage4.1.weight", "module.decoder_stage4.2.weight", "module.decoder_stage4.2.bias", "module.decoder_stage4.3.weight", "module.down_conv1.0.weight", "module.down_conv1.0.bias", "module.down_conv1.1.weight", "module.down_conv2.0.weight", "module.down_conv2.0.bias", "module.down_conv2.1.weight", "module.down_conv3.0.weight", "module.down_conv3.0.bias", "module.down_conv3.1.weight", "module.down_conv4.0.weight", "module.down_conv4.0.bias", "module.down_conv4.1.weight", "module.up_conv2.0.weight", "module.up_conv2.0.bias", "module.up_conv2.1.weight", "module.up_conv3.0.weight", "module.up_conv3.0.bias", "module.up_conv3.1.weight", "module.up_conv4.0.weight", "module.up_conv4.0.bias", "module.up_conv4.1.weight".
Unexpected key(s) in state_dict: "module.encoder1.weight", "module.encoder1.bias", "module.encoder2.weight", "module.encoder2.bias", "module.encoder3.weight", "module.encoder3.bias", "module.encoder4.weight", "module.encoder4.bias", "module.decoder2.weight", "module.decoder2.bias", "module.decoder3.weight", "module.decoder3.bias", "module.decoder4.weight", "module.decoder4.bias", "module.decoder5.weight", "module.decoder5.bias".
size mismatch for module.map4.0.weight: copying a param with shape torch.Size([2, 2, 1, 1, 1]) from checkpoint, the shape in current model is torch.Size([2, 32, 1, 1, 1]).
在上采样的时候出现这个问题是什么情况呢，是我的版本不对吗
File "/home/qinxueke/.conda/envs/vic/lib/python3.6/site-packages/torch/nn/modules/upsampling.py", line 125, in init
self.scale_factor = float(scale_factor) if scale_factor else None
TypeError: float() argument must be a string or a number, not 'tuple'
我要训练我自己的数据，为MRI格式，需要大量修改数据预处理的代码和训练代码吗？
When I load the dataset......
unable to handle huge data using single gpu how to fix this
Hi,
module 'tensorflow' has no attribute 'Summary'
How should I do for it?
Hi, thank you for sharing.
I follow your tutorial to train a model with two .nii samples.
It trained success and could be tested, but I noticed one thing that the shape of the model output didn't match with the input's whether I use raw data or fixed data as input data.
I use LiTS dataset number 28 data for testing.
Here is the code:
checkDimension.py
import SimpleITK as sitk
from scipy import ndimage
def sitk_read_raw(img_path, resize_scale=1): # 读取3D图像并resale（因为一般医学图像并不是标准的[1,1,1]scale）
nda = sitk.ReadImage(img_path)
if nda is None:
raise TypeError("input img is None!!!")
nda = sitk.GetArrayFromImage(nda) # channel first
nda=ndimage.zoom(nda,[resize_scale,resize_scale,resize_scale],order=0) #rescale
return nda
if __name__ == '__main__':
input_name = './fixed_data/data/volume-28.nii'
data_np = sitk_read_raw(input_name)
print('model input shape: {:}'.format(data_np.shape))
output_name = './output/model2/result/result-28.nii'
data_np = sitk_read_raw(output_name)
print('model output shape: {:}'.format(data_np.shape))
raw_data = './raw_dataset/LiTS_batch2/data/volume-28.nii'
data_np = sitk_read_raw(raw_data)
print('raw data shape: {:}'.format(data_np.shape))
raw_label = './raw_dataset/LiTS_batch2/label/segmentation-28.nii'
data_np = sitk_read_raw(raw_label)
print('raw data shape: {:}'.format(data_np.shape))The result is:
model input shape: (122, 512, 512)
model output shape: (64, 256, 256)
raw data shape: (129, 512, 512)
raw label shape: (129, 512, 512)
I want to visualize it with ITK-SNAP.
The original data could be visualized like this:
But currently, the output shape of the model can't be visualized like above:
I still can't find out where the code I should edit to meet my needs. Could you give me some advice? Thank you!
test.py里面变量名字有问题，predict_one_img()返回的不应该包括"test_dice"，对照train.py修改。否则会AttributeError: 'collections.OrderedDict' object has no attribute 'avg'
首先，非常感谢作者提供了开源代码！
在处理数据的时候，有语句是将三维的数据用unsqueeze(0)增加了一个维度，不是很理解这么做的意义。希望有人能解答一下！！非常感谢！
Can you show me your result when submitting LITS on leaderboard?
Thankyou very much!
有哪位大佬知道这个问题怎么解决
dice += (pred[:,i] * target[:,il)sum(din=l)sum(din=l)sun(din=l) / ((pred[:,i] * target[:,i)sum(din=l)sum(din=l).sumn(din=l)+
RuntimeError: The size of tensor a (1024) must match the size of tensor b (512) at non-sinaleton dimension 3
您好，最新更新的文件，我下载之后运行train说找不到module.nn文件，无法运行怎么回事？
Hi, thank you for your amazing work.
I thing there is some problem for dice loss, especially when u calculate with weight class.
Regrading this paper:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8393549/
the weight should be apply to both numerator and de numerator.
I have try your dice loss code and loss give negative value. I thing this problem cause by the values of numerator is too high.
I'm trying to run the code, but during training is giving me the error NameError: name 'get_dices' is not defined. Could you please provide help me to resolve this issue?
你好，我用您的模型跑了一下只分割肝脏的，结果loss降到0.3左右就不再下降了，dice1也只能达到0.7多，请问您自己能跑到0.9多不是用的这个网络吗？
这个代码我跑了四次，最后发现是自己对题主的意思理解错误导致，然后我花了几天时间逐行阅读代码，希望我的遭遇能够帮助到各位：`
①不要预处理‘测试集(test)’！！！不要预处理‘测试集(test)’！！！不要预处理‘测试集(test)’！！！重要的事情说三遍，只需要预处理训练集(train)，dataset_lits_test.py会对测试(test)集图像进行处理。
②前面前辈说到的ResUNet.py里135-161行的map里的upsample里的scale参数后两位减半。
I ran this code four times and finally realized that it was due to my incorrect understanding of the author's meaning. I spent a few days reading the code line by line, hoping that my experience can help everyone:
① Do not preprocess the 'test set'!!! Do not preprocess the 'test set'!!! Do not preprocess the 'test set'!!! Say important things three times, just preprocess the training set, dataset_ Lits_ Test. py will process the test set images.
② The last two digits of the scale parameter in the upsample in the map from lines 135 to 161 of ResUNet. py mentioned by the previous generation are halved.
③ Uncomment the 25th line of test.py. The purpose of this line depends on the annotation. Once it is done, it will be ready.We don't need to restore the parameters halved mentioned in ② when running test. py.
Hello, do you have a trained model that I can use directly
用自己的数据集train dice一直在0.02左右徘徊是为什么？
hello, when i run python ./preprocess_LiTS.py in the fix_data() function the following error occured "ERROR (nifti_image_write_hdr_img2): cannot open output file './fixed_datadata/volume-52.nii' ** ERROR (nifti_image_write_hdr_img2): cannot open output file './fixed_datalabel/segmentation-52.nii.gz'",
如題，我目前是匯入腦影像與分割結果(class=87)，我目前只有修改config.py中
parser.add_argument('--n_labels', type=int, default=87,help='number of classes')
若要開始訓練則會出現以下錯誤
File "train.py", line 89, in <module> train_log = train(model, train_loader, optimizer, loss, args.n_labels, alpha) File "train.py", line 43, in train target = common.to_one_hot_3d(target,n_labels) File "/home/ma/3DUNet-Pytorch/utils/common.py", line 9, in to_one_hot_3d one_hot = torch.zeros(n, n_classes, s, h, w).scatter_(1, tensor.view(n, 1, s, h, w), 1) RuntimeError: index 87 is out of bounds for dimension 1 with size 87
尊敬的作者您好，我想问一下，当我复现您的代码一直到训练的时候，我发现我的终端虽然会显示训练的进度条和epoch，但是过了一段时间后仅仅是显示但是并没有任何进展，这个是因为该模型训练一次epoch需要很长时间呢还是哪里出了问题？期待您的回复
你好，想问一下，在运行train.py时有一些输出不是很明白，train dice0,1,2分别是代表什么呢？同理，val dice0,1,2又分别代表什么，如果要评估分割效果，真正的dice系数是看哪一个，谢谢！
HI, @lee-zq
I met this error when run with my data.
My data is brain MRI OASIS data (nii file format).
What's wrong to me?
...
)
(map1): Sequential(
(0): Conv3d(256, 2, kernel_size=(1, 1, 1), stride=(1, 1, 1))
(1): Upsample(scale_factor=(8.0, 8.0, 8.0), mode=trilinear)
(2): Softmax(dim=1)
)
)
Total number of parameters: 9498744
=======Epoch:1=======lr:0.0001
0%| | 0/3 [00:00<?, ?it/s]torch.Size([2, 1, 48, 128, 128]) torch.Size([2, 48, 128, 128])
0%| | 0/3 [00:01<?, ?it/s]
Traceback (most recent call last):
File "/Users/tessor2/3DUNet-Pytorch/train.py", line 94, in
train_log = train(model, train_loader, optimizer, loss, args.n_labels, alpha)
File "/Users/tessor2/3DUNet-Pytorch/train.py", line 46, in train
target = common.to_one_hot_3d(target,n_labels)
File "/Users/tessor2/3DUNet-Pytorch/utils/common.py", line 9, in to_one_hot_3d
one_hot = torch.zeros(n, n_classes, s, h, w).scatter_(1, tensor.view(n, 1, s, h, w), 1)
RuntimeError: index 243 is out of bounds for dimension 1 with size 2
E
Thanks in advance ~
Best,
@bemoregt.
你好lee-zq，
非常感谢你的方法！
我在训练的时候遇到了这个问题:
=======Epoch:1=======
0%| | 0/2 [00:00<?, ?it/s]/home/jiaxi/anaconda3/envs/DKFZ/lib/python3.6/site-packages/scipy/ndimage/interpolation.py:583: UserWarning: From scipy 0.13.0, the output shape of zoom() is calculated with round() instead of int() - for these inputs the size of the returned array has changed.
"the returned array has changed.", UserWarning)
0%| | 0/2 [00:04<?, ?it/s]
Traceback (most recent call last):
File "train.py", line 103, in <module>
train_log = train(model, train_loader, optimizer)
File "train.py", line 66, in train
optimizer.step()
File "/home/jiaxi/anaconda3/envs/DKFZ/lib/python3.6/site-packages/torch/autograd/grad_mode.py", line 26, in decorate_context
return func(*args, **kwargs)
File "/home/jiaxi/anaconda3/envs/DKFZ/lib/python3.6/site-packages/torch/optim/adam.py", line 119, in step
group['eps']
File "/home/jiaxi/anaconda3/envs/DKFZ/lib/python3.6/site-packages/torch/optim/functional.py", line 94, in adam
denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(eps)
RuntimeError: CUDA error: an illegal memory access was encountered
非常感谢你的帮助。
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.