VaseMuseum: Digital Intelligent Museum for Ancient Greek Pottery

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Digital Museums & Cultural Heritage · Depth: Expert, extended

Summary

VaseMuseum is a lightweight, modular multimodal agent framework for intelligent digital museums, specifically for ancient Greek pottery. It addresses challenges in vision-language models (VLMs) for cultural heritage, such as unreliable knowledge retrieval and overconfident answers. The system integrates an interactive virtual museum with VaseAgent, supporting 2D images and 3D artifacts through multimodal perception, 3D-aware reasoning, and external knowledge retrieval from authoritative sources. Crucially, it features inference-time reliability controls: source-level control filters verifiable evidence before generation, and response-level control checks generated claims, encouraging neutral, evidence-bounded answers when support is insufficient or conflicting. A training-free GRPO-style selection mechanism further enhances reliability. Experiments in a digital museum simulation show VaseMuseum improves citation validity, reduces hallucinations on knowledge-intensive queries, and produces more neutral answers under ambiguity compared to search-enabled VLM baselines. The evaluation set includes 3,000 ancient Greek vase images, with 518 linked to the LIMC database.

Key takeaway

For Machine Learning Engineers building trustworthy multimodal systems for digital museums, you should integrate inference-time reliability controls. Implement source-level filtering for external knowledge to ensure verifiable evidence. Additionally, apply response-level control to calibrate VLM answers, preventing hallucinations and promoting neutrality when evidence is ambiguous. This approach significantly improves citation validity and groundedness, crucial for cultural heritage applications where factual accuracy and cautious interpretation are paramount. Consider a GRPO-style selector for enhanced reliability.

Key insights

VaseMuseum enhances VLM trustworthiness in digital cultural heritage via inference-time evidence control and uncertainty calibration.

Principles

• Ground VLM answers in authoritative external knowledge.
• Calibrate VLM confidence based on evidence sufficiency.
• Filter retrieved sources for validity and diversity.

Method

VaseAgent uses a DeepResearch-style tool loop for knowledge acquisition, applying source control to filter evidence and response control to audit claims against the evidence pool, optionally using a GRPO-style selector for reranking.

In practice

• Implement source control to filter unreliable web search results.
• Apply response control to ensure VLM answers are evidence-bounded.
• Use GRPO-style selection for improved reliability with multiple trajectories.

Topics

• Digital Museums
• Vision-Language Models
• Cultural Heritage
• Knowledge Retrieval
• Reliability Control
• Ancient Greek Pottery

Code references

• AIGeeksGroup/VaseMuseum

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• VaseMuseum: Digital Intelligent Museum for Ancient Greek Pottery — VaseMuseum enhances VLM reliability in digital cultural heritage by integrating external knowledge and inference-time confidence control.
• Using street view images and visual LLMs to predict heritage values for governance support: Risks, ethics, and policy implications — Visual LLMs analyzing street view images can predict building heritage values for governance, but raise significant risks regarding transparency and error.
• Visual Reasoning Benchmark: Evaluating Multimodal LLMs on Classroom-Authentic Visual Problems from Primary Education — MLLMs struggle with dynamic visual reasoning despite proficiency in static visual tasks, posing risks for educational applications.
• Preserving cultural heritage: Inside Google DeepMind’s collaboration with Pelé — AI can authentically reconstruct unfilmed historical events for cultural preservation and education.
• On the Cultural Anachronism and Temporal Reasoning in Vision Language Models — VLMs misinterpret historical artifacts due to "cultural anachronism," applying temporally inappropriate modern concepts.
• If These Walls Could Talk: Critical Play with Large Language Models in Museums — Museum LLM chatbots should embrace unreliability for critical play, using fictional characters to represent history engagingly.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.