Research map / Image generation / Autoregressive and latent generation
MapDreamer: Aerial Imagery Conditioned Latent Diffusion For Lane Level Map Generation
Julian Brandes, Philipp Crocoll, Wolfram Burgard
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ghost lane latents to the lane cardinality module (ablation) | MapDreamer · UrbanLaneGraph local tiles (1.2 m) | TOPO1 F1: 0.526 -> 0.562 | Improved |
| add lane cardinality module over fixed 64-slot padding (ablation) | MapDreamer · UrbanLaneGraph local tiles (1.2 m) | GEO1 precision: 0.413 -> 0.609 | Improved |
| add VAE latent space for diffusion instead of raw vector coordinates (ablation) | MapDreamer · UrbanLaneGraph local tiles (1.2 m) | GEO1 F1: 0.519 -> 0.620 | Improved |
| add VAE latent space for diffusion instead of raw vector coordinates (ablation) | MapDreamer · UrbanLaneGraph local tiles (1.2 m) | TOPO1 F1: 0.418 -> 0.526 | Improved |
| use 20 DDIM steps instead of 100 DDPM steps without retraining (ablation) | MapDreamer default · UrbanLaneGraph local tiles (1.2 m) | GEO1 F1 / runtime: 0.64 -> 0.62 GEO1 F1; 326 ms -> 73 ms | Mixed |
| reduce LDM hidden dimension from 2048 to 512 (ablation) | MapDreamer · UrbanLaneGraph local tiles (1.2 m) | GEO1 F1: 0.64 -> 0.59 | Mixed |
| replace BGFormer with MapDreamer latent diffusion | lane graph prediction from aerial images · UrbanLaneGraph global tiles (1.2 m) | TOPO1 F1: 0.189 -> 0.401 | Improved |
| MapDreamer vs LaneGNN given GT start poses | global lane graph · UrbanLaneGraph global tiles (2 m) | GEO2 F1: 0.642 -> 0.720 | Improved |
Closest papers by meaning
- LaneDiffusion++: Feature-Level Generative Priors for Structured Lane Graph Learning (TPAMI 2026 · similarity 0.90)
- Generative Lane Topology Reasoning via Autoregressive Model with Geometry Prior (ECCV 2026 · similarity 0.89)
- Bezier Everywhere All at Once: Learning Drivable Lanes as Bezier Graphs (CVPR 2024 · similarity 0.89)
- LaneSegNet: Map Learning with Lane Segment Perception for Autonomous Driving (ICLR 2024 · similarity 0.87)
- Controllable Road Marking Generation (NeurIPS 2026 · similarity 0.86)
- Outdoor Scene Extrapolation with Hierarchical Generative Cellular Automata (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
