Rethinking Driving World Model as Synthetic Data Generator for Perception Tasks

arXiv 25.10 2025 67.5 method, application

TLDR

Dream4Drive uses a driving world model to generate synthetic multi-view videos, enhancing downstream perception tasks and corner case detection in autonomous driving.

Reasoning

The paper addresses a gap in evaluating driving world models for downstream perception, introducing a flexible framework for generating corner cases. Strengths include a novel 3D-aware generation pipeline and a new dataset, but the abstract lacks quantitative results and comparisons to baselines.

Read-first score

Read-first score 67.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 40.

Methodology quality 25%

100

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

Recency 8%

86.7

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

Topical relevance 42%

57.1

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=dataset,github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 213.

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

Rethinking driving world models as synthetic data generators for downstream perception tasks, not just generation quality and controllability.
Dream4Drive framework that decomposes input video into 3D-aware guidance maps, renders 3D assets, and fine-tunes world model to produce edited multi-view photorealistic videos.
Ability to generate multi-view corner cases at scale for autonomous driving perception.
Contribution of DriveObj3D, a large-scale 3D asset dataset covering typical driving categories for diverse 3D-aware video editing.

Methodology

Dream4Drive first decomposes the input video into several 3D-aware guidance maps, then renders 3D assets onto these guidance maps. Finally, the driving world model is fine-tuned to produce edited, multi-view photorealistic videos, which are used to train downstream perception models. The framework is evaluated through comprehensive experiments on downstream perception tasks under various training epochs.

Key Results

Dream4Drive effectively boosts the performance of downstream perception models under various training epochs, particularly for corner case perception in autonomous driving.

Resources

Paper Project