ObjectStream: Latent Objects as Memory Anchors for Streaming Video Understanding

Source

Source name: arXiv
Source kind: paper
Authors / organization: Mingkang Dong, Muxin Pu, Jie Li, Bohan Guo, Songruo Chen, Bin Ren, Xu Zheng, Chen Zhao, Tianwen Qian, Mohamed Elhoseiny, Yuqian Fu
Venue: arXiv: cs.CV
Published: 2026-07-30
Collected: 2026-07-31
URL: http://arxiv.org/abs/2607.28312v1
PDF: https://arxiv.org/pdf/2607.28312v1
DOI: UNKNOWN

Plain-Language Brief

The paper introduces ObjectStream, a framework that uses latent objects as memory anchors for understanding streaming videos. The framework does not require external object detectors or segmentation models and allows Video Large Language Models to reason over object identities, interactions, and state changes while preserving three types of evidence: persistent object histories, transient object changes, and recent visual context.

What the Document Says

The paper presents ObjectStream, a framework for streaming video understanding.
ObjectStream uses latent objects as memory anchors.
It does not require external object detectors or segmentation models.
Video Large Language Models can reason over object identities, interactions, and state changes using ObjectStream.
Three types of evidence are preserved: persistent object histories, transient object changes, and recent visual context.

Key Concepts Mentioned

streaming video understanding
latent objects
memory anchors
Video Large Language Models
persistent object histories
transient object changes
recent visual context

Claims Made by the Document

ObjectStream improves Qwen2.5-VL-7B by 10.0 points on OVO-Bench Real-Time Visual Perception in online streaming evaluation.
On offline long-video benchmarks, ObjectStream surpasses the full-token baseline while discarding 82.5% of visual tokens.

Evidence Used

Extensive experiments on online streaming and offline long-video benchmarks

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Streaming video understanding requires models to continuously retain useful visual evidence before future questions are known. Existing approaches primarily manage the growing visual context according to token importance, temporal redundancy, or segment-level relevance, but rarely organize evidence around objects that persist and evolve over time. Thus, in this paper, we introduce ObjectStream, a training-free framework that treats latent objects as memory anchors for streaming video understanding. ObjectStream induces spatially coherent latent objects directly from frozen Video-LLM representations, links them across frames into persistent anchors, and maintains their histories under a bounded memory budget, without requiring external object detectors or segmentation models. Built on these anchors, ObjectStream preserves three complementary forms of evidence: persistent object histories, transient object changes, and recent visual context. This design enables existing Video Large Language Models (Video-LLMs) to reason over object identities, interactions, and state changes while leaving the underlying model unchanged. Extensive experiments on online streaming and offline long-video benchmarks demonstrate both effectiveness and efficiency. In online streaming evaluation, ObjectStream improves Qwen2.5-VL-7B by 10.0 points on OVO-Bench Real-Time Visual Perception, while reducing peak GPU mem-ory and TTFT by approximately 50%. On offline long-video benchmarks, it surpasses the full-token baseline while discarding 82.5% of visual tokens. These results highlight latent objects as a practical and effective organizing principle for compact streaming video memory.