Overview

Audiovisual archives are an essential resource for preserving and transmitting cultural memory. However, the rapid growth of digital video collections makes manual annotation and cataloguing increasingly difficult. MUCES addresses this challenge by developing advanced AI models that can automatically analyze videos, generate semantic descriptions, and support effective search through natural language and visual examples.

The project focuses on multimodal approaches that connect video, images, and language, enabling more accurate indexing and retrieval of complex and domain-specific content. In particular, MUCES investigates methods that can adapt to long-tail concepts, including entities, places, and events that are especially relevant to the Italian cultural heritage domain.

Objectives

• Content understanding and enrichment: Develop AI models capable of extracting semantic information from videos and images,
  enabling automatic description, tagging, and structured indexing.
• Adaptation to long-tail concepts: Design methods that recognize rare, fine-grained, and culturally specific concepts
  even when only limited training data are available.
• Large-scale retrieval and browsing: Create efficient retrieval systems that allow users to explore large audiovisual
  archives through natural language queries and example-based search.

Scientific Vision

MUCES lies at the intersection of computer vision, natural language processing, and multimedia retrieval. The project combines these research areas to develop new tools for the automatic understanding and exploration of audiovisual archives.

A core ambition is to move beyond generic AI models and toward systems capable of understanding domain-specific knowledge, particularly related to Italian cultural heritage and historical audiovisual archives.