The ACDC data set seems to be normalized such that the values are in range [-1, 1]? How did you preprocess the data? I would like to visualize the segmentation performance, but for that I would like to understand how you preprocessed it in the first place. I did not find the information in your paper, nor in this repository or in other issues.

I am doing my final year project and try to use the Synapse Dataset for testing. However, the TransUNet code framework is hard to reuse for other network. I met some problems on Synapse Dataset training and testing. I see your code and find it's more understandable and I wonder if you can give me the whole code for reference. Thanks!

Hi, i have some questiones about ACDC:

1. why can do validation and test with the similar way of training? cause i saw you use batch size=1, and "test_single_volume" procedure
2. can you share the version of test that outputs the values of LV MYO RV?

thanks a lot. your work help me a lot!

非常好的工作，感谢你的发布。我使用了自己的数据集，但在评估时出现了这个错误，我很疑惑，我的image的形状大小是[3,224,224]， label是[224,224], 我的是二分类，但出现了下面维度不匹配的情况，为什么要用 for ind in range(imag.shape)这个代码吗？这样输入的input不就变成了[1,1,224,224]吗？这和我训练时的输入形状[12,3,224,224]岂不是不匹配，预测会不会因此有问题呢？

Hi i am a starter of this field, and trying to visualize your results saved (such as case0001_pred.nii.gz) for prediction and gt, just like your paper.
I cost 2 days to figure out but failed. could you please send me the code of visualizition? many thanks in advance.

Hi,

I really appreciate your work. In the paper, it is mentioned that you re-trained the models such that you can compare your results to the previous published work. I would be really interested into getting my hands on that code.

作者您好，感谢您在本篇论文里的工作。 但是我看了论文和代码有一些疑惑，就是论文里的External Attention，代码里这部分也是对当前批次的特征x进行一些nn.Linear操作： 1.请问论文里说的该模块可以跨样本学习，是指的当前batch的样本吗？ 2.是怎么体现或者是实现跨样本的呢？ 期待您的回复，谢谢。

I tried to reproduce your code on acdc, and found that the validation set always performed better in the early stage (around 50 epoch) during the training process, and always performed worse in the later epoch, which made me not sure which model to choose. How can I improve to find the most accurate model faster?

非常感谢你们很好的工作！但我有两个问题： 问题一：我使用的是冠脉数据集。 1）如果将同一组数据集既作为训练集，也作为测试集，dice是正常的，效果还可以。 2）如果训练集和测试集使用不同的数据集，dice一直为0 3）如果训练集使用数据A、B、C、D，测试集使用A、E，那么A的dice会有一定数值，但会保持一个定值，E的dice一直为0，如下图： 问题二： 我是用您提供的ACDC的数据集，为什么avg会一直是一个定值，不会改变？ 非常期待您的回复，非常感谢！

Dear author: It's very fortunate to read your article "MIXED TRANSFORMER U-NET FOR MEDICAL IMAGE SEGMENTATION". I have a lot of interest in the work of your paper.If available, could you send me your preprocessed ACDC dataset and the complete code?My email is [email protected] you very much!

I think this is a bug in the train_mtunet_ACDC.py code. Could you example it? 88 val_outputs = model(val_image_batch) 89 val_outputs = torch.argmax(torch.softmax(outputs, dim=1), dim=1).squeeze(0) It should be: 88 val_outputs = model(val_image_batch) 89 val_outputs = torch.argmax(torch.softmax(val_outputs , dim=1), dim=1).squeeze(0)