Accelerating 3D Generation with Training-free Hash-based Feature Reuse

Hash3D, a versatile and training-free acceleration method, leverages feature redundancy in diffusion-based 3D generation to substantially reduce computational costs without compromising visual quality.