Privacy-Preserving Semantic Segmentation Using Vision Transformer