Distill3R: A Pipeline for Democratizing 3D Foundation Models on Commodity Hardware

Brandon Leblanc, Charalambos Poullis · 2026

Read Distill3R: A Pipeline for Democratizing 3D Foundation Models on Commodity Hardware by Brandon Leblanc, Charalambos Poullis (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-2602-00865.txt.

Description

Brandon Leblanc and colleagues offer a 72M student cutting size 9x, speeding answers 5x, and keeping structural consistency versus a 650M teacher. They distill 3D foundation models into compact students taught on a single PC. Their Distill3R method uses offline caching to split heavy teacher work from training through compressed supervision. A confidence-aware distillation loss uses teacher uncertainty to enable commodity-hardware training. The student trains in under 3 days on one PC, where the teacher needs GPU clusters for up to a week.

Catalog record

CREATORBrandon Leblanc, Charalambos Poullis
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41001
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00865

Browse the full library · About the Tower of Babel