Audiovisual Hack-a-thon : Exploring Methods and Data through Inclusive Collaboration
| Authors | |
|---|---|
| Year of publication | 2025 |
| Type | Workshop |
| MU Faculty or unit | |
| Citation | |
| Description | Audiovisual media plays a central role in shaping social and cultural perceptions, attitudes, and behaviors. The sheer volume of audiovisual materials produced globally every second is staggering, and the number of digitized heritage materials is also rapidly increasing. Several initiatives to study moving images computationally have emerged in recent years (Chávez Heras et al. 2023; Arnold & Tilton 2023; Chávez Heras 2024; Oiva et al. 2024; Carrive et al. 2021; Masson et al. 2020). Despite these developments, deeper integration of knowledge across disciplines, methods, tools, and infrastructures remains a significant challenge. Computational research on audiovisual materials is largely scattered across different institutions and scholars working on different corners of the disciplinary continuum. In comparison to DH research in textual materials, research on audiovisual data lags behind due to the challenges of developing and learning computational methods for analyzing multiple non-linguistic modalities, as well as difficulties accessing data sources, managing file formats and sizes, and navigating copyright restrictions. Due to these problems, starting computational research of audiovisual data is perhaps more difficult than with other types of materials. To support a deeper integration and promote the development of computational research of audiovisual materials, we propose to organize a ‘humanities hack-a-thon’ format mini-conference at the DH2025 conference. The main idea of the Audiovisual Hack-a-thon: Exploring Methods and Data through Inclusive Collaboration is to bring together scholars working with audiovisual materials in different disciplinary domains to share their knowledge and to learn from each other. We aim to unite researchers who possess data, have method expertise, and those new to the field. This composition allows the workshop to bring together scholars who have data but less experience with the latest computational methods and experts developing methods who want to test their cutting-edge techniques. Including in the group people interested in working with audiovisual data, but who do not yet have data or methods, opens the door for beginners to get more hands-on experience. Working together is also a good way for networking among people interested in audiovisual materials and fostering further collaborations. Participants can either bring data and research questions or come with a method they would like to apply. We also invite participants who do not have data or a specific method in mind, but who want to gain hands-on experience to appreciate what computational analysis of audiovisual materials could be. We will, for example, apply the BVQA (Batch Visual Question Answering) tool developed at the King’s College DH lab to a collection of Soviet era newsreel data (1944-1992). Another candidate activity is to provide interested participants with a corpus of metadata and pose estimation, 3D location and action recognition data derived from more than 50 full-length recordings of theatrical performances from Stanford University’s MIME (Machine Intelligence for Motion Exegesis) project, data which has fewer copyright encumbrances than the source recordings. Providing some packaged software tools and data corpora enables participants who wish to do so to move past many of the time-consuming and technically arduous steps of configuring software, wrangingling corpora and extracting data, ensuring that groups can start working on research questions quickly. To this end, participants bringing their own data will be encouraged to supply data that does not require cleaning during the workshop, and the organizers will create in advance Jupyter notebooks that include the data that we propose to work on. |
| Related projects: |