Bard VADER:Bard as a Virtual Assistant Device for Expert art Recognition Nowadays, generalized access to the internet enables artists to share their work with the world. This results in a rising interest in art and culture, and increases people's curiosity about the greatest art masters in history. Consequently, museums and art galleries have seen an increment in their attendance. However, these places have a limited number of personal guides, and the generally provided information through booklets or audioguides is sometimes not enough to answer customers' inquiries. Thus, there is a need for more flexible, readily-available systems that are able to provide customized information and adapt to the requirements of each individual. In this context, we propose Bard VADER, an AI-based virtual museum guide capable of detecting a piece of art using computer vision methods and retrieving information about the work using a state-of-the-art Large Language Model (LLM) like Bard.Our system workflow is divided into different stages. First, the art piece is detected using a computer vision algorithm that matches the features extracted from the image with a database of the pieces showcased at the museum/gallery. Then, when the work has been identified, we ask the LLM about information from the detected piece, such as its author, epoch, art style, or historical context. Finally, the user is able to interact with the LLM and ask about any specific topic that he/she is interested in, providing a personalized experience for each customer.This system can be implemented either in an embedded system like a Khadas Edge2 platform, allowing for each customer to have all the information in their own hands, or embodied in a robot platform that serves as a guide and is able to provide a tour through all the place. The advantage of our proposal is that it provides a neverending source of information that adapts to each individuals interests, making the experience of visiting a museum or an art gallery more interactive, engaging, and satisfying than ever before. In addition, our solution is relatively low cost, since it only requires a camera, a processing unit, and connection to an online server to retrieve the required information from the LLM.
Comments