Picture through Shutterstock
Although the iPhone is Live Photos automatically captures instant moments before and after you take a picture. Researchers at the University of Washington’s new in-depth learning algorithm don’t need them – it just “predicts” what will happen in these cases.
The team research, Introduced in detail at the Computer Vision and Pattern Recognition Conference on June 22, artificial intelligence can create realistic, high-resolution looping videos from a single photo.
Which sets the method apart from other film and animation tools is that it needs nothing more than a photo to get started. “It does not require user information or additional information” explain Aleksander Hołyński, PhD student in computer science at Paul G. Allen. He adds, “It really requires you to predict the future. And in the real world, there are almost limitless possibilities for what can happen next.”
Hołyński and his team baited the neural network with thousands of clips depicting flowing movements, including videos of waterfalls, clouds, fire, rivers, and oceans, to train it to predict the movements of images. They showed it the first frame of the video and got the algorithm to guess the motion of the clip and compare its response to the whole video so that it could refine its predictions.
Rotate the image video can be aided by a method called “splashing,” which involves moving pixels in an image to track predicted movements – but this idea sounds better on paper. “Think of a flowing waterfall,” Hołyński describes. “If you only move the pixels to the waterfall, after a few videos you won’t have any pixels at the top!”
To replace for the missing pixels, the team also included a “symmetrical crack”, which asks artificial intelligence to predict the “past” of the image as well.
Because of its data set the tool is best able to create “flowing” videos of parts of nature such as waterfalls, rivers, smoke and clouds. Over time, scientists hope to move forward with animations of other objects, such as the movement of hair in the wind.