Research map / Vision-language and multimodal / Multimodal LLMs
GeoBayes: Probabilistic Image Geo-Localization Inference via Sequential Bayesian Updating
Weimin Shi, Xiang Li, Kaige Li, Junhao Fang, Qiang Zhou, Qichuan Geng, Zhong Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes GeoBayes, a sequential Bayesian updating framework that reduces confirmation bias in MLLM-based image geo-localization by probabilistically combining visual clues into location inference.
Problem
MLLMs for image geo-localization suffer from confirmation bias, committing early to possibly incorrect location guesses driven by visual clues with varied geographic likelihoods.
Approach
Treats geo-localization as sequential Bayesian updating over candidate locations, so that evidence from successive visual clues updates a posterior rather than locking in an early guess; described in the abstract as a no-training (training-free) approach.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace CoT with GeoBayes sequential Bayesian updating | Qwen2.5-VL-7B · Im2GPS3K | accuracy within 25km (%): 31.0 -> 34.7 | Improved |
| add visual objects (MCoT) to CoT (ablation) | Qwen2.5-VL-7B · Im2GPS3K | accuracy within 25km (%): 750km 70.4 -> 71.7; 25km 31.0 -> 26.3 | Mixed |
| add image-search RAG to MCoT (ablation) | Qwen2.5-VL-7B · Im2GPS3K | accuracy within 25km (%): 26.3 -> 25.8 | Hurt |
| calibrate likelihood with discrete support rating and confidence (ablation) | GeoBayes · Im2GPS3K | accuracy within 25km (%): 31.5 -> 31.9 | Improved |
| add centering term ln2(c-3) to likelihood (ablation) | GeoBayes · Im2GPS3K | accuracy within 25km (%): 31.9 -> 34.7 | Improved |
| add hierarchical transition (ablation) | GeoBayes · Im2GPS3K | accuracy within 25km (%): 31.5 -> 34.7 | Improved |
| add Enhance web search retrieval (ablation) | GeoBayes · Im2GPS3K | accuracy within 25km (%): 31.3 -> 34.7 | Improved |
| add Replace strategy (ablation) | GeoBayes · Im2GPS3K | accuracy within 25km (%): 33.2 -> 34.7 | Improved |
| add Backtrack strategy (ablation) | GeoBayes · Im2GPS3K | accuracy within 25km (%): 33.7 -> 34.7 | Improved |
| apply GeoBayes to Qwen2-VL-7B | Qwen2-VL-7B · Im2GPS3K | accuracy within 25km (%): 29.9 -> 31.7 | Improved |
| apply GeoBayes to Qwen2.5-VL-7B | Qwen2.5-VL-7B · YFCC4K | accuracy within 1km (%): 2.1 -> 4.9 | Improved |
| GeoBayes (training-free MLLM) versus classification/retrieval methods at street level | Qwen2.5-VL-7B · Im2GPS3K | accuracy within 1km (%): 14.1 (GeoCLIP) vs 6.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Georeasoner — GeoReasoner: Geo-localization with Reasoning in Street Views using a Large Vision-Language Model (ICML 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GRE Suite: Geo-localization Inference via Fine-Tuned Vision-Language Models and Enhanced Reasoning Chains (NeurIPS 2025 · similarity 0.92)
- Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models (NeurIPS 2025 · similarity 0.92)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.91)
- GeoReasoner: Geo-localization with Reasoning in Street Views using a Large Vision-Language Model (ICML 2024 · similarity 0.91)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.89)
- ERGeoBench: A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models (ICML 2026 · similarity 0.89)
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.
