A novel coarse-to-fine framework with hierarchical neural representation is proposed to efficiently generate high-quality stylized novel views from sparse input scenes.
