SymRegFlow: Symmetry-Regularized Flow Matching for Video World Models

arXiv 2026 35.5 method, theory

TLDR

SymRegFlow applies symmetry-regularized flow matching to generate multi-view-consistent videos across continuous camera poses, achieving state-of-the-art FVD on nuScenes without novel-view supervision.

Reasoning

The paper introduces a novel regularization method for flow-matching video world models, with theoretical guarantees and strong empirical results on driving datasets. Its main limitation is the narrow scope: it focuses on camera-pose generalization rather than interactive or general world dynamics, so broader world-model claims are not fully supported.

Read-first score

Read-first score 35.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 41.

Recency 5%

100

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

Methodology quality 15%

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

Topical relevance 25%

58.6

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 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

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

Citation velocity 10%

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: 324.

Keyword Scores

video world model

world model

generative world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

SymRegFlow: a symmetry-regularized flow-matching framework for multi-view-consistent video generation across continuous viewpoints without ground-truth novel-view RGB supervision.
Per-target-pose geometric warping of source views into noisy anchors combined with masked dual-anchor supervision and cross-anchor denoising-output consistency to mitigate anchor-specific errors.
Theoretical proof under an affine Gaussian surrogate that suitable consistency regularization recovers the clean-reference optimum at fixed noise levels, strictly outperforming single- and merged-anchor baselines.

Methodology

SymRegFlow builds on flow-matching-based multi-view video world models by warping source views into noisy anchors for each target pose and applying masked dual-anchor supervision with cross-anchor denoising-output consistency. It is evaluated on Cosmos-Drive-Dreams and nuScenes using FVD, FVMD, FID, and instance preservation metrics, and is compared against single-anchor, merged-anchor, and other evaluated baselines.

Key Results

On nuScenes, SymRegFlow achieves the lowest FVD and FVMD among evaluated baselines, reducing FVD by over 31% relative to the best baseline, and source-conditioned inference attains the best FID and instance preservation. Experiments on Cosmos-Drive-Dreams and nuScenes demonstrate high-quality, multi-view-consistent autonomous-driving video generation.

Resources

Paper