Reproduce “Scaling Up Occupancy-centric Driving Scene Generation: Dataset and Method” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bohan Li, Xin Jin, Hu Zhu, Hongsi Liu, Ruikai Li, Jiazhe Guo, Kaiwen Cai, Chao Ma and 4 more
UniScenev2 introduces a scalable framework for 4D dynamic driving scene generation, producing a large-scale semantic occupancy dataset, Nuplan-Occ, and synthesizing high-fidelity multi-modal data including 3D semantic occupancy, multi-view videos, and LiDAR point clouds. The framework demonstrates improved performance in downstream autonomous driving tasks like perception and planning, driven by its hierarchical, occupancy-centric approach and novel modality-bridging techniques.
Key insights
- A large-scale, high-fidelity 3D semantic occupancy dataset is fundamental for advancing scalable, occupancy-centric driving scene generation and improving downstream AD tasks.
- Decomposing large-scale scene synthesis into hierarchical steps, starting with 4D dynamic occupancy generation, provides robust and consistent conditional guidance for multi-modal data output.
- Explicitly addressing sensor-specific properties and calibration noise through techniques like Unscented Transform and sensor-specific embeddings significantly enhances the realism and robustness of generated multi-sensor data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.