⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

UniMLVG: Unified Framework for Multi-view Long Video Generation with Comprehensive Control Capabilities for Autonomous Driving

arXiv 24.12 2024 45.5 method, system

TLDR

UniMLVG generates long, multi-view driving videos with precise control using a DiT-based diffusion model, achieving significant FID and FVD improvements.

Reasoning

The paper presents a novel framework for multi-view video generation in autonomous driving, with strengths in explicit viewpoint modeling and multi-stage training. However, it is domain-specific and does not address interactive or dynamic world modeling beyond video generation.

Read-first score

Read-first score 45.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 10.

Recency 8%
75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Reproducibility 25%
73

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

Methodology quality 25%
60

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

Topical relevance 42%
14.3

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

Reproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 313.

Keyword Scores

video world model
4
world model
2
generative world model
2
world simulator
1
world dynamics prediction
1
interactive world model
0
model-based reinforcement learning world model
0

Deep Analysis

Innovations

  • Unified framework for multi-view long video generation with comprehensive control capabilities
  • Integration of single- and multi-view driving videos into training data
  • DiT-based diffusion model with cross-frame and cross-view modules across three-stage training with multiple objectives
  • Explicit viewpoint modeling approach for multi-view video generation to improve motion transition consistency
  • Capability to handle various input reference formats (text, images, or video) and condition constraints (3D bounding boxes or frame-level text descriptions)

Methodology

UniMLVG employs a DiT-based diffusion model enhanced with cross-frame and cross-view modules, trained in three stages using both single- and multi-view driving videos. The framework incorporates explicit viewpoint modeling for consistent motion transitions and supports multiple input formats and condition constraints (e.g., 3D bounding boxes, frame-level text). Evaluation is performed against state-of-the-art models using FID and FVD metrics.

Key Results

Compared to the best models with similar capabilities, UniMLVG achieves improvements of 48.2% in FID and 35.2% in FVD, demonstrating significant gains in visual quality and temporal consistency.

Resources

PaperProjectCode

Tags