GI-PIP: Do We Require Impractical Auxiliary Dataset for Gradient Inversion Attacks?