Research map / Vision-language and multimodal / Multimodal LLMs

DialogueVPR: Towards Conversational Visual Place Recognition

CVPR 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 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.

Introduces Dialogue Place Recognition (DlgPR), where a CLIP-based multi-level retriever (CMPL) and a Qwen2.5-VL-7B questioner (DQ-pilot, SFT then GRPO) iteratively ask clarifying questions to localize places, with a new DQ-cities benchmark.

Problem

Static one-shot language-guided place recognition fails when user descriptions are ambiguous or incomplete; the paper targets interactive, multi-turn localization via clarifying questions.

Approach

Casts place recognition as dialogue: a questioner MLLM generates information-gain questions over retrieved candidates, trained with SFT on curated dialogues then GRPO using a Positional Retrieval Gain reward; difficulty of samples is scored by a Discriminative Difficulty Index for curriculum sampling.

Key results

• DlgQuest (SFT+GRPO) reaches R@1 58.4 at round 5 for LosAngeles, versus 43.2 for Qwen2.5-VL-7B at round 5 and 51.2 for PlugIR.
• Retriever CMPL achieves R@1 71.9 on LosAngeles long-description retrieval, above FG-CLIP at 56.1.
• BRI of 1.18 for DlgQuest (SFT+GRPO) versus 1.41 for PlugIR and 1.58 for Qwen2.5-VL-7B.

Limitations

• Evaluation limited to five cities from DQ-cities built on GSV-Cities imagery.
• Dialogue answers appear to be simulated from generated descriptions rather than real users.
• Teacher dialogues depend on GPT-4o generation, with only brief manual inspection of borderline cases.

Experimental setup

Datasets: GSV-Cities DQ-cities-20k DQ-cities-10k

Benchmarks: DlgQuest-Cities

Metrics: Recall@1 Recall@5 PRG BRI

Baselines: Qwen2.5-VL-7B Qwen2.5-VL-72B PlugIR CLIP Long-CLIP FG-CLIP Flair DlgQuest (SFT)

Builds on: GRPO PlugIR FG-CLIP GSV-Cities YOLO-World

Closest papers by meaning

• Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models (NeurIPS 2025 · similarity 0.88)
• GRE Suite: Geo-localization Inference via Fine-Tuned Vision-Language Models and Enhanced Reasoning Chains (NeurIPS 2025 · similarity 0.88)
• GeoReasoner: Geo-localization with Reasoning in Street Views using a Large Vision-Language Model (ICML 2024 · similarity 0.87)
• ERGeoBench: A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models (ICML 2026 · similarity 0.87)
• Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.87)
• Learning to Localize Objects Improves Spatial Reasoning in Visual-LLMs (CVPR 2024 · similarity 0.87)