Autoregressive latent generator turns multi-view images into 3D Gaussian scenes

kwangmoo_yi · x · 2026-09-05

Lucas and Pietrantoni et al. present "Sparse auto-regressive modeling for scene generation from multi-view images": multi-view images plus pointmaps feed an autoregressive latent-space generator, decoded as 3D Gaussians for scene generation.

Related event: SPAR3S Generates Full 3D Scenes from Sparse Views Without 3D Supervision(2 posts)→

Original post →

More from Multimodal

Multimodal channel →