InfiniCube: Unbounded and Controllable Dynamic 3D Driving Scene Generation with World-Guided Video Models

ICCV 25 2025 54.1 method, application

TLDR

InfiniCube generates unbounded, controllable dynamic 3D driving scenes using a map-conditioned voxel world and grounded video models.

Reasoning

The paper presents a novel method combining sparse-voxel 3D generation with video models for high-fidelity, controllable driving scenes. Strengths include scalability and controllability; weaknesses are potential complexity and lack of explicit real-world dataset mention in the abstract.

Read-first score

Read-first score 54.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 41.

Recency 8%

86.7

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

Methodology quality 25%

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

Topical relevance 42%

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 25%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 415.

Keyword Scores

video world model

generative world model

world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Map-conditioned sparse-voxel-based 3D generative model for unbounded voxel world generation
Repurposing a video model with pixel-aligned guidance buffers grounded on the voxel world for consistent appearance synthesis
Fast feed-forward approach using voxel and pixel branches to lift dynamic videos to dynamic 3D Gaussians with controllable objects

Methodology

InfiniCube constructs a map-conditioned sparse-voxel-based 3D generative model to generate unbounded voxel worlds. It then repurposes a video model by grounding it on the voxel world through pixel-aligned guidance buffers to synthesize consistent appearances. Finally, a fast feed-forward approach with voxel and pixel branches lifts the dynamic videos to dynamic 3D Gaussians, enabling control via HD maps, vehicle bounding boxes, and text descriptions.

Key Results

The method generates controllable and realistic 3D driving scenes, and extensive experiments validate its effectiveness and superiority over previous approaches.

Resources

Paper Project