📝 Our paper on improving robustness of Segment Anything Model is accepted at NeurIPS 2025.