Cycle Consistency in Video Object-Centric Learning

Rongzhen Zhao, Zhiyuan Li, Ruonan Wei, Juho Kannala, Joni Pajarinen · 2026

Read Cycle Consistency in Video Object-Centric Learning by Rongzhen Zhao, Zhiyuan Li, Ruonan Wei, Juho Kannala, Joni Pajarinen (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-2605-30211.txt.

Description

Rongzhen Zhao and four co-authors tackle cycle consistency in self-supervised video object-centric learning. Models must discover objects and track them over time, unlike tracking from given detections, and slots are stochastic and unclear. Explicit cycle consistency on latent slots forces rigid mean seeking, harming equally valid decompositions and driving feature collapse. Their Implicit Cycle Consistency shifts the constraint to the reconstruction manifold, letting slots reach soft consensus rather than forcing exact alignment. Experiments on video benchmarks show it avoids feature collapse and beats explicit baselines.

Catalog record

CREATORRongzhen Zhao, Zhiyuan Li, Ruonan Wei, Juho Kannala, Joni Pajarinen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34404
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.30211

Browse the full library · About the Tower of Babel