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

OccSora: 4D Occupancy Generation Models as World Simulators for Autonomous Driving

arXiv 24.5 2024 72.3 method, application

TLDR

OccSora is a diffusion-based 4D occupancy generation model that simulates driving scenes as a world simulator for autonomous driving.

Reasoning

The paper introduces a novel diffusion-based approach for long-term 4D occupancy generation, addressing inefficiencies of autoregressive models. Strengths include trajectory-conditioned generation and temporal consistency, but it is limited to occupancy representation and evaluation on a single dataset (nuScenes).

Read-first score

Read-first score 72.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 50.

Reproducibility 25%
85

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

Recency 8%
75.1

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

Topical relevance 42%
71.4

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

Methodology quality 25%
60

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

Field roles

Reproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 84.

Keyword Scores

world simulator
10
world model
9
generative world model
8
world dynamics prediction
8
video world model
7
model-based reinforcement learning world model
5
interactive world model
3

Deep Analysis

Innovations

  • Diffusion-based 4D occupancy generation model for autonomous driving world simulation
  • 4D scene tokenizer for compact discrete spatial-temporal representations of long-sequence occupancy videos
  • Trajectory-prompt conditioned generation enabling controllable 4D occupancy video synthesis
  • Ability to generate 16-second videos with authentic 3D layout and temporal consistency

Methodology

OccSora employs a 4D scene tokenizer to compress 4D occupancy input into compact discrete spatial-temporal tokens, enabling high-quality reconstruction of long-sequence occupancy videos. A diffusion transformer is then trained on these tokens to generate 4D occupancy conditioned on a trajectory prompt. The model is evaluated on the nuScenes dataset with Occ3D occupancy annotations.

Key Results

OccSora generates 16-second occupancy videos with authentic 3D layout and temporal consistency, demonstrating its ability to understand spatial and temporal distributions of driving scenes.

Resources

PaperCode

Tags