DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation

CVPR 25 2025 58.1 method, application

TLDR

DriveDreamer4D uses world models as data machines to synthesize trajectory videos, enhancing 4D driving scene representation with significant FID improvements.

Reasoning

The paper introduces a novel approach leveraging world model priors for 4D reconstruction, addressing limitations in complex driving maneuvers. Strengths include clear problem motivation and quantitative gains, but the abstract lacks details on world model architecture and training, and generalizability is unclear.

Read-first score

Read-first score 58.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 36.

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=baseline,experiment,metric,result

Topical relevance 42%

51.4

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=github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 299.

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

Using world model as a data machine to synthesize novel trajectory videos for 4D driving scene representation
Explicitly leveraging structured conditions to control spatial-temporal consistency of traffic elements in generated videos
Proposing a cousin data training strategy to merge real and synthetic data for optimizing 4D Gaussian Splatting (4DGS)
First to utilize video generation models for improving 4D reconstruction in driving scenarios

Methodology

DriveDreamer4D employs a world model to generate synthetic videos of novel driving trajectories, where structured conditions enforce spatial-temporal consistency of traffic elements. A cousin data training strategy is then used to combine real and synthetic data for optimizing a 4D Gaussian Splatting (4DGS) representation. The approach is evaluated on novel trajectory views against baselines including PVG, S3Gaussian, and Deformable-GS.

Key Results

DriveDreamer4D achieves relative improvements in FID of 32.1%, 46.4%, and 16.3% over PVG, S3Gaussian, and Deformable-GS respectively, and relative increases in NTA-IoU of 22.6%, 43.5%, and 15.6%, with a user study confirming enhanced spatiotemporal coherence.

Resources

Paper Project