Process-of-Thought Reasoning for Videos

Jusheng Zhang, Kaitong Cai, Jian Wang, Yongsen Zheng, Kwok-Yan Lam, Keze Wang · 2026

Read Process-of-Thought Reasoning for Videos by Jusheng Zhang, Kaitong Cai, Jian Wang, Yongsen Zheng, Kwok-Yan Lam, Keze Wang (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-2602-07689.txt.

Description

Video understanding needs temporally grounded, multi-step thinking over long, noisy observations, not just recognition of what appears. Jusheng Zhang and five co-authors propose Process-of-Thought reasoning for videos, which structures inference into lightweight, verifiable steps.

The method interleaves time evidence choice, step-wise state updates, and constrained answer synthesis, refining hypotheses while keeping traceability to video evidence. It plugs into current vision-language backbones for closed-book or tool-augmented thinking. A unified trace representation aligns decisions with time segments, improving robustness to distractors. Video thinking teams get an explicit, checkable thinking process instead of a single opaque answer.

Catalog record

CREATORJusheng Zhang, Kaitong Cai, Jian Wang, Yongsen Zheng, Kwok-Yan Lam, Keze Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE86 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07689

Browse the full library · About the Tower of Babel