⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

OmniWorld: A Multi-Domain and Multi-Modal Dataset for 4D World Modeling

arXiv 25.9 2025 60.8 benchmark

TLDR

OmniWorld is a large-scale, multi-domain, multi-modal dataset for 4D world modeling, enabling benchmarks and improving SOTA in reconstruction and video generation.

Reasoning

The paper's strength lies in its comprehensive dataset addressing data scarcity in 4D world modeling, with clear performance gains demonstrated. However, the abstract lacks explicit real-world data (dataset appears synthetic/game-based) and does not cover interactive or RL-based world models, limiting scope.

Read-first score

Read-first score 60.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 33.

Methodology quality 25%
90

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark,dataset,evaluation,experiment,metric

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%
47.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%
46

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset,github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 366.

Keyword Scores

world model
9
video world model
7
generative world model
6
world dynamics prediction
5
world simulator
3
interactive world model
2
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • Introduction of OmniWorld, a large-scale multi-domain multi-modal dataset for 4D world modeling
  • OmniWorld-Game dataset with richer modality coverage, larger scale, and more realistic dynamic interactions compared to existing synthetic datasets
  • Establishment of a challenging benchmark that exposes limitations of current SOTA approaches in 4D world modeling
  • Demonstration that fine-tuning SOTA methods on OmniWorld yields significant performance gains in 4D reconstruction and video generation

Methodology

OmniWorld comprises a newly collected OmniWorld-Game dataset and several curated public datasets spanning diverse domains. The dataset is designed to provide richer modality coverage, larger scale, and more realistic dynamic interactions than existing synthetic datasets. A benchmark is established to evaluate SOTA methods, and fine-tuning experiments are conducted on existing SOTA approaches for 4D reconstruction and video generation tasks.

Key Results

Fine-tuning existing SOTA methods on OmniWorld leads to significant performance gains across 4D reconstruction and video generation tasks. The benchmark also reveals limitations of current SOTA approaches in modeling complex 4D environments.

Resources

PaperProject

Tags