A Surreal World, the objects of YouTube Shorts
Stefan Schär and Andrin Brülhart, A Surreal World, the objects of YouTube Shorts, 2024
YouTube is the most popular video streaming network in the world. A large portion of the content on YouTube are so called YouTube Shorts. These Videos are very short, in most cases under 30 seconds. This very fast-paced, short-lived environment demands for a new form of content. This content varies from attention-grabbing to outright bizarre. The viewers on the platform reward this bizarre, strange content with their most valuable currency: clicks and attention.
The content of the videos was extracted using a script which collected the most trending Youtube Shorts at the time and an object detection model.
The object detection model cocossd – COCO stands for Common Object in Context; it is a large-scale object detection, segmentation, and captioning dataset – recognises the objects in the video and groups them in one of 80 categories. These categories vary from person, cars and animals to mundane all-day objects. Looking at the dataset, the categorization appears very object-based fitting well to the very commercial, shiny capitalist world of YouTube Shorts content generators.
Every time the model detects something from its dataset it extracts a frame of the video. With these frames new short videos were created with a sorted and grouped view on just the detected objects. The grouped display of the objects shows popular images in the world of YouTube Shorts. This limited, newly arranged content brings an understanding of the surreal world of YouTube Shorts. Drawing inspiration from the work of Sam Lavigne, it also highlights the mistakes and false understandings of the objects-detection model.