Fast direct multi-person radiance fields from sparse input with dense pose priors

João Paulo Lima, Hideaki Uchiyama, Diego Thomas, Veronica Teichrieb

Abstract: Highlights•We propose a novel approach for multi-person radiance field recovery.•We deal with sparse single-frame input without previous multi-view person matching.•We exploit dense pose priors for person segmentation and radiance field supervision.•We gain time for optimization while reaching higher accuracy in synthesized images.