Research map / Vision-language and multimodal / CLIP and zero-shot
AddressCLIP: Empowering Vision-Language Models for City-wide Image Address Localization
Shixiong Xu, Chenghao Zhang, Lubin Fan, Gaofeng Meng, SHIMING XIANG, Jieping Ye
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 image-caption contrastive loss on top of address loss (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 77.66 -> 79.20 | Improved |
| add image-geography matching loss on top of address loss (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 77.66 -> 79.27 | Improved |
| combine caption and geography losses with address loss (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 77.66 -> 80.39 | Improved |
| train with caption loss only, without address loss (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 77.66 -> 69.27 | Hurt |
| unfreeze both image and text encoders instead of text only (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 48.88 -> 80.39 | Improved |
| unfreeze text encoder in addition to image encoder (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 77.77 -> 80.39 | Improved |
| append address text to scene caption (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 79.32 -> 80.39 | Improved |
| use BLIP-Caption-Large instead of Base for scene captions (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 80.28 -> 80.39 | No clear effect |
| reduce images per location from 24 to 3 (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 80.39 -> 56.86 | Hurt |
| train on mixed Pittsburgh + San Francisco data, test on SF (ablation) | AddressCLIP · SF-IAL-Base | SSA-1 (%): 86.32 -> 85.51 | Hurt |
| restrict candidate addresses to prior streets (W=2) (ablation) | AddressCLIP · Pitts-IAL | SSA-1 (%): 80.39 -> 89.57 | Improved |
| end-to-end AddressCLIP vs Image-GPS-Address retrieval pipeline | AddressCLIP · Pitts-IAL | SSA-1 (%): 75.17 -> 77.01 | Improved |
Closest papers by meaning
- Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models (NeurIPS 2025 · similarity 0.86)
- GeoReasoner: Geo-localization with Reasoning in Street Views using a Large Vision-Language Model (ICML 2024 · similarity 0.85)
- Pixel-Aligned Language Model (CVPR 2024 · similarity 0.84)
- Textual Supervision Enhances Geospatial Representations in Vision-Language Models (ICML 2026 · similarity 0.84)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.83)
- ResCLIP: Residual Attention for Training-free Dense Vision-language Inference (CVPR 2025 · similarity 0.83)
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.
