A novel 3D generation method is introduced, enabling versatile and high-quality 3D asset creation. At its core is a unified Structured LATent (SLAT) representation, which supports decoding into multiple output formats, including Radiance Fields, 3D Gaussians, and meshes. This is achieved by combining a sparsely-populated 3D grid with dense multiview visual features extracted from a robust vision foundation model, effectively capturing both structural (geometry) and textural (appearance) information while preserving decoding flexibility.
Laurent Lescop – 3D Viz – Parametric Design – Immersion – AI // Recherches et réalisations de Laurent Lescop en visualisation 3D, conception paramétrique, immersion et IA appliquées à l'architecture.
Structured 3D Latents for Scalable and Versatile 3D Generation (https://trellis3d.github.io/) Continuer la lecture de IA-project workflow part10 
