I think mosaic or mixup need MultiImageDataset.

Because it requires mix_results for other mixed image.

All I have to do is move MultiImageDataset?
If not, could you advise me how to adjust it to the semantic segmentation task?

Or just add mosaic transform first?