MapDreamer generates lane-level maps from aerial images using diffusion models

Researchers have developed MapDreamer, a novel generative diffusion model capable of synthesizing lane-level vector maps directly from single aerial images. This model utilizes a compact latent representation for lane centerlines and their topological relationships, predicted by a transformer-based latent diffusion model. MapDreamer conditions its denoising steps on aerial features via cross-attention to ensure alignment with the observed scene and incorporates a lane cardinality module with background ghost lane latents to manage varying lane counts. Experiments on the UrbanLaneGraph dataset, derived from Argoverse 2, demonstrate superior geometric and topological fidelity compared to existing non-generative methods. AI

IMPACT This research could accelerate the creation of high-definition maps for autonomous driving, potentially reducing manual labor and improving map accuracy.

RANK_REASON This is a research paper detailing a new model for map generation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Argoverse 2
• MapDreamer
• UrbanLaneGraph

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →