📝 Daniel's Notes


Search IconIcon to open search

day two

Last updated Jul 17, 2023

# 7 July

# Fast prototyping session

On day two, we gathered again to decide what ideas we could put to the test and fast-prototype. We examined two datasets: Mila’s soviet newsreels and this publicly available sample from BBC Archives. We decided to focus on the newsreels, which have richer metadata. The idea was to test two methods that map to our discussions of historical time and meaningful objects, and cinematic time and fast/slow editing. We tested on the newsreels:

We know these techniques have limitations, as explored during day one of the workshop. But we also think that, combined, these methods complement each other and can be useful for the analysis of temporal dynamics in large collections of moving images, for example as a way to detect editing patterns and co-relate them to themes and known historical time frames.

Testing this combination against the newsreel data set is helpful because thanks to Mila’s sustained work on these reels, we know something about their local as well as global structures, so we can apply the methods to the collection of images and use the collection of images as a way to test the the methods. This approach reminded me of the notion of “ predicting the past” by the wonderful Tobias Blanke.

During the panel on high-dimensional cinema, I referred to this back and forth between modalities of inquiry as building “aesthetically sensitive systems”, and in our discussion the day after Nanne suggested “aesthetically sensitive retrieval” as an applied version of this idea. We wrapped up the day with some ideas and next steps along these lines.

Mila and Andrea also sampled the news reels and produced an annotation guidelines document for possible future human annotation, and that directly informs editing style detection and open CLIP queries.

Thanks to James Graham and Matt Penn from the e-research team at KCL, who helped us setup ad-hoc VMs to run these tests in the CREATE HPC cluster from the Digital Humanities Computer Lab (pictured above).