Triangle for a Box, Single Channel HD, 5’47”, 2014

Triangle for a Box looked at the latent normativity in detection algorithms and how they could, in turn, change an actor’s relationship to their performance. The editing treats gesture and language as separate tracks that can fall in and out of phase. This project was the first in which I used the Upper-body Configuration Detector by The Visual Geometry Lab at Oxford. After learning how the program analysed images, I wrote a script to throw this process into relief. 

The ‘triangle’ in the title refers to the rotating cast of characters around the table. The moment they black box their interaction, by occluding bodies (Step 5) is a strategy whereby the same script can be performed to in subtly different ways to confound machine vision. The hope here was to make the cast of the film set a kind of foil to the learning process in the machine. In practical terms, we see the red bounding boxes of the detection algorithm being confused by the images shot. This, in turn, points back to the dataset the algorithm was trained with: hundreds of episodes of largely American sitcoms. The shoot became a study in projection, foregrounding a disconnect between these bodies and those in the dataset. —2015