Meta-acervos is an experimental visual research platform for finding and comparing works from different art collections through their formal and intrinsic characteristics, such as colour, composition and the presence of figurative elements.
Alongside traditional metadata such as author, title and date, the system uses information extracted directly from images of the works, proposing an alternative way to read collections and establish relationships between them.
The tool was developed by the Acervos Digitais e Pesquisa Thematic Project team and is available as open-source software.
Meta-acervos begins with the challenge of thinking about digital collections as a network: how can data scattered across different platforms be brought together to allow new connections between collections that rarely meet?
In its first version, the system brings together 4,200 works, mainly paintings and drawings, from 17 Brazilian collections with open, structured data, including Wikimedia GLAM, Brasiliana Museus and MAC-USP.
The images are standardized and analysed using multimodal vision and language models. The results are numerical vectors, or embeddings, that describe each work’s visual content and make it possible to identify affinities.
Produced by: Acervos Digitais e Pesquisa – Fapesp
Coordination: Giselle Beiguelman
Programming: Thiago Hersan, João Generoso
Design: Luisa Rodrigues
Review: Ana G. Magalhães, Bruna Keese, Cássia Hosni and Renata Perim
