HPC for Cultural Heritage

HPC for Cultural Heritage

Since the end of 2022, the pre-Exascale supercomputer Leonardo has joined Cineca’s other supercomputers in supporting scientific research and Cultural Heritage projects, offering further interesting possibilities. There are two fields in particular in which supercomputing can make significant contributions to Cultural Heritage: the analysis and management of Big Data and the training of artificial intelligence networks, as well as their application to large volumes of data.

Cineca has already exploited supercomputing in projects such as I-Media-cities, which collected audiovisual resources dedicated to the urban context from nine European archives. Artificial intelligence algorithms were used to work on the audiovisual material, automatically identifying video shots, cutting them and metadating them with the use of object detection systems. The enrichment of metadata, applied to the individual shot, greatly facilitates the subsequent work of researchers who, for example, may wish to identify specific iconographic elements useful for their research, while avoiding the long hours of checking audiovisual content, in which they are not even sure of finding what they are looking for.

The I-Media-Cities platform, judged as market ready by the European Innovation Radar, has already been applied in another project, DARE Digital Environment for Collaborative Alliances to Regenerate Urban Ecosystem in middle-sized cities, but without the inclusion of AI algorithms. However, the flexibility of the platform, which is available in an open format, allows the inclusion of any algorithm suitable for AI analysis. Processing data using the algorithms deemed most appropriate and in a short time is precisely what makes HPC support indispensable for projects such as this.

The training of neural networks is another extremely useful and promising area for the world of Cultural Heritage. The automatic tagging of audiovisual digital resources, as in the I-Media-Cities project, is based, as mentioned, on the object detection process. Algorithms, in order to detect and recognise objects, must be trained. That is to say, they must be subjected to very large datasets with examples of what is then to be recognised automatically by the algorithm.

Currently, the networks are mainly trained on contemporary content. However, the heritage of Cultural Heritage does not stop at the last thirty or forty years, thus making the already trained networks available difficult to use. Hence the importance of training new networks suitable for content that is more meaningful for the Cultural Heritage domains.

In addition to these two fields, there is a third one with increasing relevance. In a context characterised by a shift towards what we might call cyberspace, with its metaverses, supercomputing resources and XR (Extended Reality) frameworks can cooperate in several ways to enhance the experience and capabilities of XR applications for Cultural Heritage as well.

First of all, supercomputing resources can be used to provide the processing power needed to perform complex simulations and models in real time within XR environments, enabling faster and more accurate rendering of dynamic scenes and interactions, thereby creating more realistic and detailed environments and interactions. In addition, XR applications often require the processing of large amounts of data, and supercomputing resources can help manage this data efficiently and effectively. They can also be used to support the creation of collaborative virtual environments, allowing multiple users to interact within the same XR environment.

By collaborating, supercomputing resources and XR frameworks can create powerful and immersive XR experiences that were previously unimaginable.