WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models
Summary
WorldMemBench 提出了一个用于系统评估视频世界模型世界记忆能力的基准,涵盖260个测试用例,分别考察静态记忆(人物、物体、环境的身份与几何保持)和动态记忆(事件演进与状态持续性)。对九个代表性模型的评估显示,强静态记忆并不意味着可靠的状态演化,且重复重访会暴露记忆退化问题。
View Cached Full Text
Cached at: 10/16/26, 07:04 AM
Paper page - WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models
Source: https://huggingface.co/papers/2610.19063 Authors:
,
,
,
,
,
,
,
,
,
,
,
Abstract
WorldMemBench evaluates world memory in video models by testing static and dynamic retention across repeated revisits and state transitions.
As video models enable longer and more realistic interactive experiences, maintaining a world consistent with its history becomes essential. This requires both preserving previously observed content and ensuring coherent world evolution over time.World memoryis a critical capability to evaluate alongside visual fidelity and video duration. We introduce WorldMemBench, a benchmark for systematically evaluatingworld memoryinvideo world models. Across 260 test cases, it assessesstatic memorythrough the retention of human identity and appearance, object appearance and geometry, and environmental content and geometry, anddynamic memorythroughevent progressionandstate persistence. Using controlled camera departures andmulti-revisit trajectories, the benchmark evaluates static content against fixed initial references and dynamic events against expected state transitions. It combinesappearance and geometry metricswith event-specific assessments, and reports video quality and camera-following accuracy as complementary diagnostics. Evaluation of nine representative models shows that strong static retention does not imply reliable dynamic state evolution, while repeated revisits reveal memory degradation beyond the first return. Human studies support the alignment between our automatic evaluations and human judgments. WorldMemBench provides a systematic testbed for diagnosing memory failures and guiding the development of persistent, temporally coherent video worlds.
View arXiv pageView PDFProject pageGitHub12Add to collection
Get this paper in your agent:
hf papers read 2610\.19063
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper1
#### rescuerz/WorldMemBench-weights Updatedabout 23 hours ago • 1
Datasets citing this paper2
#### rescuerz/WorldMemBench Viewer• Updatedabout 23 hours ago • 23.4k • 3.1k • 5 #### rescuerz/WorldMemBench-examples Viewer• Updatedabout 23 hours ago • 20 • 133 • 1
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.19063 in a Space README.md to link it from this page.
Collections including this paper1
Similar Articles
MBench: A Comprehensive Benchmark on Memory Capability for Video World Models
This paper introduces MBench, a benchmark for evaluating the memory capabilities of video world models across entity, environment, and causal consistency over long temporal horizons.
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
MemoBench is a diagnostic benchmark for evaluating video generation models' memory consistency in dynamically changing environments, where objects disappear and reappear in updated states. It includes 360 ground-truth clips and an evaluation suite combining automated metrics with VQA-based assessment, revealing insights into memory consistency challenges.
MemoWM: How World Models Change What Agents Need to Remember
MemoWM 提出一种以世界模型为条件的记忆分配框架,通过共享预测压缩智能体长期经验存储,在五个长时程智能体记忆基准上达到42.42%平均准确率,同时将经验特定存储降低53.9%。研究还揭示了共享模型容量与重复存储成本之间的权衡关系。
More Memory, Better Videos? A Controlled Study of Visual Memory in Long-Video Generation
本文提出 VMem-Bench,一个两阶段诊断基准,系统研究长视频生成中视觉记忆(检索历史画面)对历史一致性与当前故事准确性的权衡影响,发现检索虽提升实体召回率却降低了精确度与禁用实体规避能力。
Addressable Memory for Video World Models
This paper introduces WorldTrace, a training-free memory framework for long-horizon video world models that keeps compressed cache addressable, plus LoopBench, a benchmark for episodic recall after long detours. It improves temporal consistency by +15.5% and episodic recall by +19.5% on LoopBench.