Published on:
28 November 2023
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Ivan Lopes,
Fabio Pizzati,
Raoul de Charette
Proposes novel task of material extraction from photos
Uses text-to-image model to generate tileable material textures
Decomposes textures into albedo, normal, roughness maps
Validation on synthetic and real-world material datasets
Showcases material editing of 3D scenes
Extracting Materials from Photos
This paper proposes a two-step method to extract realistic 3D materials from regions in a single photo, without needing multi-view captures or 3D geometry. First, a text-to-image model generates tileable texture images resembling the region's material. Then a domain-adapted neural network decomposes textures into reusable albedo, normal, roughness maps. Experiments validate the quality of extracted materials and showcase editing of 3D scenes.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper