PanoWorld: Real-World Panoramic Generation

arXiv 2026 44.4 method, benchmark

TLDR

PanoWorld uses rotation-equivariant properties for long-range memory in panoramic world models, with a new dataset World360.

Reasoning

The paper introduces a novel approach leveraging rotation-equivariance for panoramic world models, supported by a new dataset combining real-world and simulated data. Strengths include the innovative use of geometric transformations and a comprehensive dataset; weaknesses are the lack of detailed quantitative results and limitations in the abstract.

Read-first score

Read-first score 44.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 37.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 18%

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

Topical relevance 29%

52.9

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 18%

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 546.

Keyword Scores

world model

generative world model

video world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Exploits rotation-equivariant property of omnidirectional representations to simplify camera trajectories into translations via fixed headings
Dense Panoramic Ray-Conditioning (DPRC) for current-action modeling
Geometry-aware Memory Augmentation (GMA) for long-range memory
Three-stage training pipeline for progressive optimization
World360 dataset: large-scale real-world and simulated panoramic clips for evaluating physical consistency under spatial variations and diverse illumination

Methodology

PanoWorld uses rotation-equivariant omnidirectional representations to model camera motion as translations with fixed headings, incorporating DPRC and GMA modules. A three-stage training pipeline progressively optimizes these components. Evaluation is performed on the newly introduced World360 dataset, which contains real drone-captured and simulated panoramic videos.

Key Results

PanoWorld outperforms alternative methods by a large margin on the World360 dataset, as shown in extensive experiments.

Resources

Paper