somet question about label value

[songwaimai]

Hello, I have changed your program, and it can run smoothly without any error. However, I have observed that the value of label does not seem to be passed in. May I ask if there is such a problem in your program.

[MinaRe]

Please check the load image sequence in utils.py, indeed each 3d Seg0,1,2 has corresponded to the segmentation region!

[songwaimai]

thanks! Actually, I use the iseg2017 dataset, maybe the format of label is different. which dataset do u use??

[MinaRe]

I applied this code for LiTS2017, ACDC 2017, BraTS 2015-2019 and ,... for semantic segmentation, in order to get a better result I would recommend changing loss_weights = [10, 1] in train.py and adding a complementary label in the training time.

[songwaimai]

Thanks very much! I will try it~

[MinaRe]

[MinaRe]

and weight must be adjusted regarding the imbalanced ratio and/or difficulties of training ....

[songwaimai]

so kind of u~ do you have published the related article? Would you mind sending me the name of your article or the link for me to learn about...

[songwaimai]

Also, I wonder if the data sets you use have multiple labels? label of iseg 2017 is only one corresponded to one T1 and one T2 of a subject, so i want to know how each to process can make 3d Seg0,1,2 has corresponded to the segmentation region?

[MinaRe]

Nice, in this case (http://iseg2017.web.unc.edu/files/2019/02/Benchmark-on-Automatic-6-month-old-Infant-Brain-Segmentation-Algorithms.pdf) I would solve the segmentation (which is not imbalanced) with probabilistic 3D GAN, which is similar to this code but you need to add 2 prior losses for T1 and T2 and dropout in the generator.