So you’d have to use an AI to guess what the scene was like in the past and the future.So the mannequin challenge was the perfect dataset.It has people in all sorts of positions, but they are all frozen in time.
Hundreds of custom-selected TikToks with different dances, indoors and outdoors, and entirely different vibes were used by a researcher to help computers learn to see. This was not to compare trendy dances, but rather to provide ground truth for AI models. This dataset was more varied and better than just using 3D scans from Renderpeople, as it included backgrounds, jackets, and different shapes. To use the TikToks, the researcher had to create a 3D picture, which was done by removing the background and estimating the movement in 3D space. Additionally, the mannequin challenge was used as a dataset, as it had people in all sorts of positions, but frozen in time. Google researchers created “A Dataset of Frozen People” using tons of mannequin challenge videos. This dataset was used to teach a model to guess what a scene will look like if a camera moves five feet away. It provided thousands of flashcards for the model to learn from, with a variety of settings, types of people, and backgrounds. These researchers used the Mannequin Challenge as ground truth to check their work. The variety in the Mannequin Challenge was invaluable as it put their program through its paces and tested it against a Carnegie Mellon dataset. The Mannequin Challenge was also in the real world, providing a full set of flashcards with answers that could be checked. The machines are learning, but for now, we are still the teachers, as they are only as good as the flashcards we provide.