Reproduce “CityGen: Structure-Guided City-Style Synthesis for Cross-City Autonomous Driving” 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
Zezhong Qian, Zhao Yang, Lu Tan, Zhihao Yan, Weiyi Hong, Haizhuang Liu, Yawei Jueluo
The CityGen framework addresses the challenge of deploying autonomous driving systems across diverse urban environments by synthesizing target-city-style data without requiring new labels. This approach, which disentangles structural layout from visual appearance, improves performance in 3D object detection, BEV semantic segmentation, and planning tasks when tested on unseen cities.
Key insights
- Disentangling scene structure (geometry, object layout) from visual appearance (texture, lighting) enables efficient synthesis of new city-style data while preserving label consistency.
- Combining precise geometric conditioning (via HD-maps) with high-level visual style descriptors (from VLMs) in a diffusion model effectively generates realistic and diverse urban scenes for cross-city adaptation.
- Rigorous evaluation on geographically disjoint benchmarks is crucial for assessing the true transferability and robustness of autonomous driving models across different urban domains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.