SLiR enables sound, tight linear relaxations of general activation functions using only Lipschitz constants or critical points. It achieves up to 7.8x more verification properties than state-of-the-art methods by efficiently computing upper and lower bounds via a shifting procedure.