Open smallboy-code opened 8 months ago
Hi How did you train this multi-class approach? It seems like you trained on a segmentation mask with a single channel, but different values (0,1,2,3) for the different classes? I suggest you train with one-hot-encoding, i.e., one binary mask for each of the classes.
Yes, I trained on a segmentation mask with a single channel with different values (0,1,2,3) for the different classes. The one binary mask for each of the classes means training on a segmentation mask with three channels with WT,TC and ET?
yes exactly, all three classes will have a separate binary channel each.
Yes, I have also try to do this, but I found the finnal output will be like this:
WT,TC,ET,respectively. I think the results are worse than one channel.
Yes, I have also try to do this, but I found the finnal output will be like this:
WT,TC,ET,respectively. I think the results are worse than one channel.
Hi, I also want to use multi-class segmentation for training, have you solved this problem? @smallboy-code
It remains unresolved. @ZhengChen6
Hi @JuliaWolleb , thank you for your great work. I want to solve the multi-class segmentation problem and I delete the "label = torch.where(label > 0, 1, 0).float()". Now, I got some output looks not well as follows:
![7](https://github.com/JuliaWolleb/Diffusion-based-Segmentation/assets/61675340/cb03b4f6-64d2-4f71-87ac-8f1825c8805b)
So, can you give me some advices about this problem?