Abstract

GAME aims to develop a deep learning generative framework with a high potential for positive impact on science, technology, individuals and society. 

The project focuses on two key aspects: 

(1) images-to-image outpainting composition

(2) 3D synthesis/image rendering with conditions and style transfer from textual records.

By recreating virtual scenes of museum artefacts and incorporating visitors as central subjects, the project aims to provide a personalized digital museum experience, while respecting historical context, culture, and object styles. 

Methodological challenges include incorporation of multi-modal data to drive content and style in scene synthesis, a natural integration of subjects into the environment, optimization of process, parameters and guidance weighting, and human-in-the-loop interaction for historical realism. 

The project importance lies in addressing limitations of existing approaches, pushing the limits of adaptability to a real-world scenario and offering innovative solutions, ultimately contributing to the advancement of ethical and inclusive generative AI models, and enhancing the understanding and enjoyment of cultural heritage in the digital landscape.