About the subsampling

[GitHub-TXZ]

Thanks for your great job. But I have some confusions. Why did you rescale it to 224? Have you unified all the samples to the same voxel spacing? Looking forward to your reply, thank you in advance.

[hitachinsk]

There are two considerations to this operation: (1) TransUnet and SwinUnet both rescale the image to 224, we just follow their data processing operations. (2) CT image contains noise, which may hinder the segmentation quality. If we rescale the CT image to a smaller resolution, the noise can be reduced to some extent.

[GitHub-TXZ]

so did you consider the voxel spacing? In nnunet, the author unify voxel spacing in the preprocessing, thank you for you reply.

[hitachinsk]

I did not consider the voxel spacing, but that is a feasible solution.

[GitHub-TXZ]

I just can't quantify the influence of voxel spacing. hahahha, I am so confused whether should I unify the voxel spacing.

[hitachinsk]

Maybe it's a direction to be explored. Thanks for your suggestions.

[hgtlab]

作者您好，我用您的数据集运行了train，保存的pth文件用于测试中，但是test的时候预测输出的全为0，可是使用您训练好的pth测试时候，就会生成预测结果，请问一下是需要该哪部分代码吗