search

yeerwen / UniSeg

MICCAI 2023 Paper (Early Acceptance)
Other
174 stars 5 forks source link

output channel #20

Closed kimsekeun closed 11 months ago

kimsekeun commented 11 months ago

Thanks for great work.

"Note that the channel number of multi-scale segmentation maps is set to the maximum number of classes among all tasks." it means that for output channel of decoder, it set to 3 (organ ,tumor, vertebra) ? If I understand correctly the final output channel is 3 for all datasets?

Screenshot 2023-12-02 at 4 34 42 PM
yeerwen commented 11 months ago

The number of output channels is set to 4 as it also includes the background class.

yeerwen commented 11 months ago

The channel number aligns with the maximum number of classes among all tasks and does not correspond to the organ, tumor, and vertebra.

kimsekeun commented 11 months ago

To clarify the classes and task, classes includes organ list and task means dataset A, B, C or is it ( MR, CT ...)?

e.g. dataset A : liver kidney hepatic colon spleen dataset B : liver lung

  1. is output channel considered as 6 + 1 (background). and tasks is 2 (dataset A, B). is that correct ?
  2. I think MOTS have 7 classes at least, but how did you set as 4 in paper?
  3. When calculating loss, for example in dataset B, if output channel is 7 then how you calculate loss in class that is not annotated in dataset b?

thank you in advance.

yeerwen commented 11 months ago

Q1: Task is 2 (A and B), output channel 5 (class number of A) + 1 (background). Q2: See Q1. Q3: Ignoring the output channels corresponding to the classes that are not annotated in dataset b.