Fractal Autoregressive Depth Estimation with Continuous Token Diffusion

Jinchang Zhang, Xinrou Kang, Guoyu Lu · 2026

Read Fractal Autoregressive Depth Estimation with Continuous Token Diffusion by Jinchang Zhang, Xinrou Kang, Guoyu Lu (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-2603-14702.txt.

Description

Zhang, Kang and Lu estimate depth from one image with a fractal autoregressive diffusion design. Autoregressive generation suits depth poorly in raw form. Pixel-wise steps waste compute, and discrete tokens jar with smooth depth values. Their method builds depth scale by scale, coarse to fine. A fusion module blends image features with the current depth guess. A diffusion loss models depth in continuous space directly. Reused generator units form a fractal stack for speed. A consensus step merges samples and flags unsure pixels.

Catalog record

CREATORJinchang Zhang, Xinrou Kang, Guoyu Lu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44548
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14702

Browse the full library · About the Tower of Babel