search

CUHK-AIM-Group / DiffRect

[MICCAI' 24] DiffRect: Latent Diffusion Label Rectification for Semi-supervised Medical Image Segmentation
MIT License
22 stars 2 forks source link

Mistake #2

Closed Xi-Mu-Yu closed 2 weeks ago

Xi-Mu-Yu commented 2 weeks ago

Hello, there seems to be an error with val.list and test.list in the MS-CMRSEG19 dataset. For example, test.list in mscmrseg19_split1 has only one file name. The test.list of mscmrseg19_split2 is the same as the val.list

xinyuliu-jeffrey commented 2 weeks ago

Hi @BeiZhiRen, thanks for your interest in our work!

Different from ACDC that have both val and test, in MS-CMRSEG19 dataset and Decathlon Prostate, we actually only split the dataset into training and validation dataset. The performance are reported as the average of performance on the validation sets of two different splits. The test list are not used for these two datasets.

Hope this helps!

Xi-Mu-Yu commented 2 weeks ago

您好，感谢您对我们的工作感兴趣！

与 ACDC 同时具有 val 和 test 不同，在 MS-CMRSEG19 数据集和迪卡侬前列腺中，我们实际上只将数据集拆分为训练数据集和验证数据集。性能报告为两个不同拆分的验证集上的平均性能。测试列表不用于这两个数据集。

希望这有帮助！ Ok, I see, thanks for your answers and contributions in semi-supervised semantic segmentation, I will do my best to quote and promote your article

xinyuliu-jeffrey commented 2 weeks ago

@BeiZhiRen Thanks again for bring this to our attention!