TFboys-lzz/MPSCL

Reproduction of the results of the paper.

XiudingCai opened this issue · 5 comments

XiudingCai commented

I found that using the provided yaml file directly only reproduces half of the results in the paper. How can I reproduce the results reported in the paper? For example, the number of optimization iterations, batch size, etc.

TFboys-lzz commented

Hi Xiuding:
Have you loaded the pre-trained model yet?

XiudingCai commented
XiudingCai commented

Yes, I followed the readme file you provided strictly and I ran it 50k times as per the YAML file provided. But the reproduced experimental results are very low (about 24% for the dice metric) for both MPSCL and ADVENT, and I don't know what went wrong.

TFboys-lzz commented

Hi Xiuding:
You can check if the images and labels are corresponding, and also check the performance of the model on the source domain

XiudingCai commented

Thanks for the quick reply, I printed the paths of the images and their corresponding labels and found that they were inconsistent, now I fixed the problem and started training again. Thanks again for your reply, it was very helpful!