First Shape, Then Meaning: Efficient Geometry and Semantics Learning for Indoor Reconstruction

Remi Chierchia, Léo Lebrat, David Ahmedt-Aristizabal, Olivier Salvado, Clinton Fookes, Rodrigo Santa Cruz · 2026

Read First Shape, Then Meaning: Efficient Geometry and Semantics Learning for Indoor Reconstruction by Remi Chierchia, Léo Lebrat, David Ahmedt-Aristizabal, Olivier Salvado, Clinton Fookes, Rodrigo Santa Cruz (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-03463.txt.

Description

Remi Chierchia, Léo Lebrat, David Ahmedt-Aristizabal, Olivier Salvado, Clinton Fookes, and Rodrigo Santa Cruz present FSTM, an indoor reconstruction method. It learns shape first and meaning second. Neural surface methods give good room geometry. But adding object semantics the usual joint way is slow and scales poorly.

A geometry warm-up optimizes the signed distance field from images and geometric cues alone. Semantic fields are estimated afterward. That simple split beat multi-surface rivals. It trained more than twice as fast on Replica. It handled real-world flaws better on ScanNet++, and recovered more object surfaces, raising recall without specialized modules.

Catalog record

CREATORRemi Chierchia, Léo Lebrat, David Ahmedt-Aristizabal, Olivier Salvado, Clinton Fookes, Rodrigo Santa Cruz
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.03463

Browse the full library · About the Tower of Babel