Research map / Vision-language and multimodal / Cross-modal learning
Pix2map: Cross-Modal Retrieval for Inferring Street Maps From Images
Xindi Wu, KwunFung Lau, Francesco Ferroni, Aljoša Ošep, Deva Ramanan
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 |
|---|---|---|---|
| Use cross-modal image-graph retrieval instead of unimodal nearest-neighbor retrieval (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 3.2168 -> 1.5908 | Improved |
| Add Chamfer-based partial credit to the contrastive training loss (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 3.2249 -> 1.7440 | Improved |
| Add edge loss to contrastive plus Chamfer training (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 1.7440 -> 1.5908 | Improved |
| Use early fusion of camera views instead of separate image encoders with average pooling (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 4.8129 -> 1.5908 | Improved |
| Resample graph vertices to be approximately equidistant (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 3.2663 -> 1.5908 | Improved |
| Restrict transformer attention using the graph adjacency mask (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 1.9309 -> 1.5908 | Improved |
| Add adjacency information as a node input feature (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 1.9241 -> 1.5908 | Improved |
| Add positional encodings to graph vertices (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 1.5908 -> 2.1564 | Hurt |
| Expand the retrieval library with unpaired maps (ablation) | Pix2Map · Argoverse MapUpdate | Chamfer: 1.5908 -> 1.5208 | Improved |
| Increase Pittsburgh retrieval library size from 5.7k to 40k graphs (ablation) | Pix2Map · Pittsburgh MapUpdate | Chamfer: 1.5908 -> 1.5208 | Improved |
| Use Pix2Map rather than TOPO-TR | Street-map inference · Argoverse MapUpdate | Chamfer: 3.0140 -> 2.0882 | Improved |
| Test on novel areas rather than previously seen regions | Pix2Map · Pittsburgh MapExpand vs MapUpdate | Chamfer: 1.5908 -> 2.6654 | Hurt |
Closest papers by meaning
- Spatial Retrieval Augmented Autonomous Driving (CVPR 2026 · similarity 0.87)
- VectorMapNet: End-to-end Vectorized HD Map Learning (ICML 2023 · similarity 0.85)
- Cross-view image geo-localization with Panorama-BEV Co-Retrieval Network (ECCV 2024 · similarity 0.85)
- 3D StreetUnveiler with Semantic-aware 2DGS - a simple baseline (ICLR 2025 · similarity 0.84)
- CVGL: Causal Learning and Geometric Topology (NeurIPS 2025 · similarity 0.84)
- UrbanIng-V2X: A Large-Scale Multi-Vehicle, Multi-Infrastructure Dataset Across Multiple Intersections for Cooperative Perception (NeurIPS 2025 · similarity 0.84)
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.
