TikTok has revolutionized the way we share and consume content. From lip-syncing to dance challenges, it’s become a platform for creativity and expression. But TikTok isn’t just about entertainment; its videos are also being used to train AI algorithms. In particular, researchers at the University of Washington have been using TikTok dances as part of their research into depth estimation—the ability for computers to understand how far away objects are from each other in an image or video.
The team was inspired by the Mannequin Challenge, which went viral on TikTok in 2020. The challenge involves people freezing in place while someone films them with a camera moving around them in a circle. This type of motion is known as “structure from motion” (SFM) and can be used to estimate depth information from images or videos taken with cameras that move around an object or scene.
To create their dataset, the team collected over 10 million frames of footage from more than 1,000 different Mannequin Challenges posted on TikTok between March and August 2020. They then manually labeled each frame according to its relative distance from the camera—close up shots were labeled “near” while further away shots were labeled “far”—and used this data to train their algorithm for depth estimation tasks such as 3D reconstruction and object tracking.
The results showed that using SFM data generated by humans performing popular dances on TikTok could improve accuracy when compared with traditional datasets created using robotic arms or drones flying around scenes capturing footage at various distances from objects within those scenes. This suggests that there may be potential applications for this type of data beyond just entertainment purposes: it could help machines better understand our environment so they can interact with us more naturally in future applications like autonomous vehicles or augmented reality systems.
This research demonstrates how creative activities like dancing can be leveraged not only for fun but also for practical purposes such as training AI algorithms — something that would have been unimaginable even just a few years ago! It shows us yet again how powerful social media platforms like TikTok can be when it comes to driving innovation and pushing boundaries across multiple disciplines including computer vision, robotics, machine learning, and artificial intelligence research fields alike!
|How TikTok Dances Trained an AI To See|AI|Vox
