Clarification on Semantic Segmentation Training on Pascal VOC

[yagmurakarken]

Thank you for your excellent work.

I noticed that the Pascal VOC semantic segmentation data loader in the repository appears to be using binary segmentation masks. Specifically, during data loading, masks corresponding to specific classes are converted into binary masks rather than multi-class masks.

Could you please clarify:

• Did you intentionally use binary segmentation masks for training semantic segmentation models on Pascal VOC as stated in the code, or was this configuration leftover from previous experiments?

• If your training setup only supports binary segmentation, how did you measure or interpret the performance on a multi-class dataset like Pascal VOC?

Any clarification would be greatly appreciated!

Thank you again for sharing your work.