Data augmentation changed the mask in semantic segmentation

so, which transform operations can I apply to my train dataset?