Stereo World Model: Camera-Guided Stereo Video Generation

arXiv 26.3 2026 57.8 method

TLDR

A camera-conditioned stereo world model for end-to-end stereo video generation with improved consistency and efficiency.

Reasoning

The paper introduces novel camera-frame RoPE and stereo-aware attention decomposition, achieving strong results on benchmarks and real-world applications like VR rendering and embodied learning. However, it lacks explicit evaluation on interactive or dynamic prediction tasks, and the term 'world model' is used loosely without full dynamics modeling.

Read-first score

Read-first score 57.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 42.

Recency 8%

100

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

Topical relevance 42%

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

Methodology quality 25%

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

Reproducibility 25%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 359.

Keyword Scores

world model

video world model

generative world model

interactive world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified camera-frame RoPE that augments latent tokens with camera-aware rotary positional encoding for relative, view- and time-consistent conditioning while preserving pretrained video priors via stable attention initialization
Stereo-aware attention decomposition that factors full 4D attention into 3D intra-view attention plus horizontal row attention, leveraging epipolar prior for disparity-aligned correspondences with lower compute

Methodology

StereoWorld is a camera-conditioned stereo world model that jointly learns appearance and binocular geometry for end-to-end stereo video generation, operating exclusively in RGB modality while grounding geometry from disparity. It uses a unified camera-frame RoPE for conditioning and a stereo-aware attention decomposition to efficiently capture disparity-aligned correspondences.

Key Results

StereoWorld improves stereo consistency, disparity accuracy, and camera-motion fidelity over monocular-then-convert pipelines, achieving more than 3x faster generation with an additional 5% gain in viewpoint consistency.

Resources

Paper Project