Knowledge NeRF: Few-shot novel view synthesis for dynamic articulated objects

Open Webpage

Wenxiao Cai, Xinyue Lei, Xinyu He, Junming Leo Chen, Yuzhi Hao, Yangang Wang

Published: 01 Jan 2025, Last Modified: 09 Nov 2025J. Vis. Commun. Image Represent. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Proposes Knowledge NeRF for few-shot novel view synthesis of dynamic objects.•Achieves high-quality 3D reconstruction with only 5 images per pose.•Leverages pretrained NeRFs as knowledge bases to guide new pose synthesis.•Matches image quality of NeRF trained with 100 images using sparse inputs.