AVD2: Accident Video Diffusion for Accident Video Description

arXiv 25.3 2025 43.7 method

TLDR

AVD2 generates accident videos with natural language descriptions to improve accident scene understanding, creating the EMM-AU dataset and achieving state-of-the-art performance.

Reasoning

The paper addresses data scarcity in accident scenarios by generating videos with descriptions, which is a strength. However, it lacks real-world validation and focuses narrowly on accident video generation without broader world model claims.

Read-first score

Read-first score 43.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 0.

Methodology quality 25%

100

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

Recency 8%

86.7

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

Reproducibility 25%

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

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 560.

Keyword Scores

world model

world simulator

generative world model

interactive world model

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Novel AVD2 framework for generating accident videos aligned with detailed natural language descriptions and reasoning
Contribution of the EMM-AU (Enhanced Multi-Modal Accident Video Understanding) dataset

Methodology

AVD2 employs a diffusion-based video generation model conditioned on natural language descriptions and reasoning to produce accident videos. The generated videos are compiled into the EMM-AU dataset, which is then used to enhance accident scene understanding. Performance is evaluated using automated metrics and human evaluations against existing baselines.

Key Results

Integration of the EMM-AU dataset achieves state-of-the-art performance on accident video understanding tasks, as measured by both automated metrics and human evaluations.

Resources

Paper Project