Dear author: I have been studying on your work for a while. I noticed that in pancreas_preprocess.py you crop all image according to label and saved them. So testing on foreground areas only. I'm confused whether it should testing on the whole image, using sliding window method. Looking forward to your reply!

Dear authors, I have been studying your article. I have a concern with Table III. The values of DSC (%) with different ratios of labeled to unlabeled samples. You use three ratios, i.e., 1:4, 1:2, 1:1. For these three ratios, did you use the same number of labeled samples or you divide the whole dataset with these ratios? For example, we have a dataset with 1000 images, with three ratios above and 100 labeled samples, the corresponding samples are: 100:400, 100:200, and 100:100. If I was wrong, please correct me. Thank you

Dear authors, Firstly, thank you for your amazing work. To reproduce your results, I used your pretrained V-Net to train semi-supervised segmentation. I first tested your pretrained V-Net on Pancreas dataset (I pre-processed this dataset before testing by your code) but the results are very different. Specifically, it achieved 0.1801 Dice, 0.1062 Jaccard, 50.0687 ASD, and 24.28 95HD while your results in the paper are 0.6996 Dice, 0.5555 Jaccard, 1.64 ASD, and 14.27 95HD. Can you explain about it or update your repository with testing file ???

HDF5-DIAG: Error detected in HDF5 (1.10.6) thread 0: #000: /tmp/SimpleITK-build/ITK/Modules/ThirdParty/HDF5/src/itkhdf5/src/H5F.c line 370 in itk_H5Fis_hdf5(): unable open file major: File accessibilty minor: Not an HDF5 file #001: /tmp/SimpleITK-build/ITK/Modules/ThirdParty/HDF5/src/itkhdf5/src/H5Fint.c line 830 in itk_H5F__is_hdf5(): unable to locate file signature major: File accessibilty minor: Not an HDF5 file #002: /tmp/SimpleITK-build/ITK/Modules/ThirdParty/HDF5/src/itkhdf5/src/H5FDint.c line 126 in itk_H5FD_locate_signature(): unable to read file signature major: Low-level I/O minor: Unable to initialize object #003: /tmp/SimpleITK-build/ITK/Modules/ThirdParty/HDF5/src/itkhdf5/src/H5FDint.c line 205 in itk_H5FD_read(): driver read request failed major: Virtual File Layer minor: Read failed #004: /tmp/SimpleITK-build/ITK/Modules/ThirdParty/HDF5/src/itkhdf5/src/H5FDsec2.c line 725 in H5FD_sec2_read(): file read failed: time = Fri Nov 26 10:42:53 2021 , filename = '/media/lab549/CE56C94660FE3E83/semantic_segmentation_xhq/code_xhq/CoraNet-master/dataset/Pancreas-CT/data/Pancreas-CT/PANCREAS_0001/11-24-2015-PANCREAS0001-Pancreas-18957/Pancreas-99667', file descriptor = 4, errno = 21, error message = 'Is a directory', buf = 0x7fff8af532a8, total read size = 8, bytes this sub-read = 8, bytes actually read = 18446744073709551615, offset = 0 major: Low-level I/O minor: Read failed Exception thrown in SimpleITK ImageFileReader_Execute: /tmp/SimpleITK/Code/IO/src/sitkImageReaderBase.cxx:105:

Thank you for your paper and code. I'm wondering which script to run to replicate the paper results with the proposed method? I ran train_ST_3D_VNet.py and achieved the baseline results (e.g., Dice of 69 for 12 labeled/50 unlabeled images on the pancreas CT dataset), but it's not clear which training script to run to get the best performance (e.g., Dice of 79 for 12 labeled/50 unlabeled images on the pancreas CT dataset).

Dear author: I noticed that in dataset/pancreas.py, dataset length of labeled & unlabeled data are different:

class Pancreas(Dataset):
...
def __len__(self):
      if self.split == 'train_lab':
         return len(self.image_list) * 5
      else:
         return len(self.image_list)
...

It's different from the setting of UA-MT and DTC, I'm curious about the reason. Looking forward to your reply!

Dear author, thank you very much for your work in this field. I would like to consult you about the missing module in the Train UA mt.py file about Train_Vnet.py. Looking forward to getting your reply~

Dear author: Appreciate to your amazing work first. I run your code on pancreas dataset, pretrain model could get 76.62% accuracy on test set. While the TABLE.Ⅴ in paper shows that the V-net accuracy is 69.96%, and I'm confused about the difference. Hope you could help me! Looking forward to your reply!