Differentiable visual prompts enhance semantic segmentation in adverse conditions, outperforming existing methods.
