InfoTok: Adaptive Discrete Video Tokenizer via Information-Theoretic Compression

ICLROral2026

Authors: Haotian Ye, Qiyuan He, Jiaqi Han, Puheng Li, Jiaojiao Fan, Zekun Hao, Fitsum Reda, Yogesh Balaji, Huayu Chen, Sheng Liu, Angela Yao, James Zou, Stefano Ermon, Haoxiang Wang, Ming-Yu Liu
Affiliation: Stanford University
Venue: ICLR 2026
Track: Oral

TL;DR

This paper introduces InfoTok, an adaptive video tokenizer guided by information theory, which significantly boosts video compression efficiency and reduces computational overhead without degrading visual quality.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

theory video

← All ICLR 2026 Oral papers · Browse the whole archive